Classification of Individual Dairy Cow Behaviors Using Accelerometer, Gyroscope, and Integrated Sensor Models

Classification of Individual Dairy Cow Behaviors Using Accelerometer, Gyroscope, and Integrated Sensor Models | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Classification of Individual Dairy Cow Behaviors Using Accelerometer, Gyroscope, and Integrated Sensor Models Khamta Pongsanun, Tadsorn Apirak, Leklerdsiriwong Aekaluck, Sanphet Chunithipaisan, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6682405/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 20 Nov, 2025 Read the published version in BMC Veterinary Research → Version 1 posted 10 You are reading this latest preprint version Abstract Background: Automated behavior monitoring is increasingly important in precision dairy farming, supporting early disease detection, welfare assessment, and productivity optimization. Although accelerometers effectively detect postural changes, they have limited capacity to capture rotational or transitional movements. Gyroscopes provide complementary angular velocity data that may enhance classification of complex behaviors. However, their combined use remains underexplored, particularly at the individual cow level. This study aims to evaluate the performance of accelerometer, gyroscope, and combined sensors models for classifying four key cow behaviors: lying, standing, eating, and walking at the individual animal level. Results: Over 780,000 labeled observations were obtained from seven dairy cows monitored over 90 days. Lying behavior consistently produced low, stable signals across all axes of accelerometer and gyroscope, while eating showed the greatest variability, particularly along the X and Y axes. Significant axis-specific and behavior-specific differences were observed (p < 0.05), with GyroY and GyroZ capturing the highest rotational activity during eating and walking. Signal vector magnitudes effectively distinguished behaviors, with lying showing the lowest values and eating the highest. Random Forest models combining accelerometer and gyroscope data consistently outperformed single-sensor approaches, particularly for classifying lying and standing behaviors. Although eating and walking exhibited lower sensitivity, sensor fusion improved classification robustness across individuals. Conclusion: The integration of accelerometer and gyroscope data enhanced classification accuracy, particularly for static behaviors. Axis-specific signal patterns and individualized modeling revealed critical insights into behavior differentiation and cow-specific variability. These findings support the development of scalable, sensor-based monitoring systems tailored to precision livestock management. Dairy cow behavior Behavior classification Precision livestock farming Random Forest machine learning Sensor fusion Tri-axial accelerometer and gyroscope Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Background Monitoring livestock behavior is a fundamental aspect of precision dairy farming, supporting animal welfare, early disease detection, and productivity improvements [1-3]. Accurate behavior monitoring facilitates the early detection of health and reproductive conditions, such as lameness, metabolic disorders, and estrus, thereby enhancing herd management outcomes [2, 4, 5]. Traditional observation methods, such as visual scoring, are labor-intensive, subjective, and limited in both temporal resolution and scalability, which restricts their utility in modern herd management systems [6]. As an alternative, wearable sensor technologies, particularly accelerometers, have emerged as reliable tools for continuous and automated behavior monitoring in cattle [7-9]. Accelerometers are particularly effective at detecting postural changes and linear motions associated with behaviors such as lying, standing, or walking [9, 10]. However, accelerometers have limited ability to detect nuanced or rotational behaviors, such as transitions and head movements, due to their sensitivity being restricted to linear acceleration [9, 11]. In contrast, gyroscopes, which measure angular velocity, provide complementary information by capturing rotational movement data and have shown potential to enhance behavior classification, particularly for complex or dynamic behaviors such as walking [9, 11, 12]. These automated systems are particularly valuable in pasture-based or labor-constrained farming environments, where continuous manual observation is often infeasible [13, 14]. With the growing adoption of individualized precision management in dairy systems, behavior classification at the individual cow level using multimodal sensor data is becoming increasingly important for enabling targeted interventions and improving health and performance [2, 15]. Numerous studies have validated the effectiveness of accelerometers in detecting fundamental cattle behaviors such as lying, standing, and eating. For instance, tri-axial accelerometers could accurately classify lying and standing behaviors, though walking and transitions were more difficult to detect [9]. Similarly, high precision and sensitivity in classifying resting and eating behaviors using neck- and leg-mounted accelerometers have been reported [16]. However, accelerometers primarily capture linear motion and may struggle to detect rotational or complex movements, potentially leading to the misclassification of behaviors such as transitions between lying and standing [9, 16]. To address these limitations, recent studies have investigated the integration of gyroscope data, which measures angular velocity, to enhance behavior classification models. Machine learning algorithms applied to collar-mounted gyroscopes have demonstrated higher classification accuracy than accelerometers alone, with classification performance reaching up to 99% for specific cattle behaviors [17]. Although these findings are promising, the standalone and integrated use of gyroscope data remains insufficiently explored under real-world farming conditions, where factors such as sensor displacement, environmental interference, and individual variability may compromise data integrity [12]. Moreover, most existing classification models aggregate data across multiple animals, potentially obscuring individual variability in movement patterns. Such aggregation can mask subtle, cow-specific behavioral patterns, ultimately reducing the precision and applicability of classification models [18, 19]. Few studies have systematically evaluated and compared the performance of accelerometer-only, gyroscope-only, and combined sensor models at the individual-cow level, revealing a significant gap in the literature. Addressing this gap is essential for advancing individual-based monitoring systems capable of accurately interpreting the unique behavioral patterns of each animal, thereby improving animal welfare and farm management practices [16]. This study aims to evaluate the performance of accelerometer, gyroscope, and combined sensor configurations for classifying cattle behaviors at the individual animal level. It addresses a critical knowledge gap by examining the distinct signal characteristics of each axis (X, Y, Z) from tri-axial accelerometers and gyroscopes in relation to four primary behaviors: lying, standing, eating, and walking. To assess the classification performance, models are developed and compared using three sensor input strategies: accelerometer-only, gyroscope-only, and a combination of both. A Random Forest classifier is employed due to its robustness in handling noisy, high-dimensional data and its demonstrated effectiveness in livestock behavior classification tasks [17, 20]. By conducting the analysis at the individual animal level rather than aggregating data across multiple animals, this research provides refined insights into movement-based behavioral classification. The findings are expected to inform the development of more precise and adaptable monitoring systems tailored to individual cattle, thereby advancing individualized livestock management. Material and Methods 1. Animal Management The study was conducted over a three-month period (August to October 2021) at the loose barn of the Farm Animal Hospital, Faculty of Veterinary Science, Chulalongkorn University, Nakhon Pathom, Thailand. Seven heifers (75–87.5% Holstein-Friesian) were housed in a 30 × 15 m concrete-floored enclosure designed to allow free movement and expression of natural behaviors. The cows were fed a total mixed ration comprising 80% roughage and 20% concentrate (on a dry matter basis) at 2.5% of body weight, provided twice daily at 09:00 and 14:00 h. Clean drinking water was available ad libitum throughout the study period. 2. Sensor Configuration and Placement Acceleration and angular velocity were continuously recorded over 90 days using a custom-built activity meter operating 24 hours per day. The device was developed in-house and comprised a tri-axis accelerometer and gyroscope sensor (MPU-6050, InvenSense Inc., California, USA), integrated with a wireless module, microcontroller, 3,700 mAh lithium battery, external antenna, and Wi-Fi router (Figure 1, left). The MPU-6050 includes an onboard digital motion processor capable of processing 6-axis motion data. The accelerometer and gyroscope featured full-scale measurement ranges of ±2, ±4, ±8, and ±16 g, and ±250, ±500, ±1000, and ±2000°/s, respectively [21]. The sensor system was supported by a LoRa mainboard (Heltec Automation, Sichuan, China) for power management and wireless data transmission. Devices were enclosed in 3D-printed housings and securely mounted on the right side of each cow’s neck using adjustable collars (Figure 1, right) [22]. Antennas were positioned within 50 meters of the sensors to ensure consistent data communication with the gateway. A dedicated computer, connected via LAN to a Wi-Fi router, served as the data collection hub. The system utilized XAMPP (Apache Friends, Berlin, Germany) for preliminary server testing and phpMyAdmin version 4.8.1 (phpMyAdmin Project, www.phpmyadmin.net) for SQL-based data management. 3. Data Processing and Analysis The overall workflow of this study comprises three main stages: data acquisition, data preparation, and classification model development. These stages represent a structured process that begins with the collection of synchronized sensor and behavioral data, continues with systematic data cleaning and feature preparation, and concludes with the development and evaluation of classification models using supervised machine learning techniques. As illustrated in Figure 2, this framework provides the basis for the detailed methodological descriptions presented in the following sections. 3.1 Data Acquisition Sensor data were continuously recorded over a 90-day period and stored in time-series format with timestamp annotations. Simultaneously, cow behaviors were recorded using a closed-circuit television (CCTV) system operating at 15 frames per second. Synchronization between video footage and sensor data was achieved through precise timestamp alignment. Two trained observers independently annotated behaviors by reviewing 24-hour video recordings throughout the 90-day study. Each time window was labeled as lying, standing, eating, or walking based on a standardized ethogram [23, 24]. To ensure consistency, both observers underwent joint training and annotated a pilot dataset to harmonize behavior interpretation. Inter-observer reliability was assessed using Cohen’s Kappa on randomly selected video segments. A 5% random subset of the dataset, sampled across individual cows and time periods, was annotated by both observers to ensure behavioral and temporal diversity. Inter-observer reliability, assessed using Cohen’s Kappa, was 0.84, indicating strong consistency. Discrepancies were resolved through discussion and consensus meetings. Annotation was conducted blind to the study’s hypotheses and model development, in line with best practices outlined in recent video-based annotation studies. Segments containing artifacts, missing values, or overlapping behaviors were excluded from analysis. Although seven behaviors were initially annotated, only four including lying, standing, eating, and walking were retained for model development. Drinking was excluded due to an insufficient number of observed instances, while ruminating and other were removed because of substantial overlap with other behaviors and ambiguous neck movements that reduced classification reliability. Consequently, the final dataset included only four clearly distinguishable and mutually exclusive behaviors, as defined in Table 1. Table 1. Definition of cow behaviors. Behavior Definition Lying The cow is in a resting posture with the ventral body surface in contact with the ground, supported by the sternum and one or both thighs. The neck is positioned either vertically or horizontally and may be flexed backward toward the hindquarters. Lateral recumbency, in which the cow lies fully on its side, was excluded from this category to maintain consistency in posture-based labeling. Standing The cow remains upright, supported by at least three legs, without forward or backward movement. The neck is aligned along the vertical axis, although minor movements related to comfort or social interactions may occur. Eating The cow stands in the feeding area on at least three legs with its head lowered into the feed bunk to ingest or masticate feed. The behavior is considered to end when the cow raises its head and maintains that position for at least five consecutive seconds. Walking The cow exhibits progressive movement, either forward or backward, covering more than two feet. The behavior involves sequential limb movements, with the head generally held in an upright position. 3.2 Data Preprocessing Sensor data preprocessing was performed using Python within a Jupyter Notebook environment. The workflow included data import, inspection, cleaning, noise filtering, and feature extraction. Prior to automated processing, raw data were manually reviewed to ensure format consistency and structural completeness. 3.2.1 Data Cleaning Each data window was designed as a fixed 10-second interval. To ensure temporal consistency, segments with substantial timestamp deviations, indicative of signal loss or transmission errors, were excluded from analysis. A threshold of ±1 second from the expected window duration was used to identify these irregular segments, following established practices [20, 25]. This filtering step minimized artifacts and improved the reliability of downstream feature extraction. The proportion of excluded segments was minimal and did not substantially affect the overall sample size or behavioral class distribution. Missing values, malformed entries, and extreme outliers were removed using null value filtering functions, consistent with best practices in wearable sensor data analysis [20, 25]. To reduce signal noise, a rolling average filter was applied to smooth the time-series data prior to feature extraction. Prior to feature extraction, the dataset was segmented according to predefined behavioral labels. This ensured that only stable, single-behavior intervals were included, minimizing the influence of behavioral transitions, and enhancing classification accuracy [25-27]. 3.2.2 Feature Extraction and Selection Following preprocessing, sensor data were segmented into 10-second windows, from which 79 descriptive features were extracted to characterize the movement patterns of dairy cows. These features included time-domain statistical metrics such as mean, standard deviation, minimum, maximum, sum, root mean square (RMS), kurtosis, interquartile range (IQR), and zero-crossing rate, as well as complexity and magnitude descriptors like entropy, signal magnitude area (SMA), and signal vector magnitude (SVM) [28, 29]. Entropy was calculated using scipy.stats.entropy, while SMA and SVM were derived from Euclidean combinations of tri-axial sensor data. Movement-specific indicators including dynamic body acceleration on individual axes (DBAX, DBAY, DBAZ), overall dynamic body acceleration (ODBA), and vectorial dynamic body acceleration (VeDBA) were computed following established formulas [30]. In total, 42 features were derived from the accelerometer and 37 from the gyroscope, utilizing both raw axis values and composite statistical measures. A detailed breakdown of feature categories and sensor-specific counts is provided in Table 2. To ensure signal continuity and feature integrity, any window containing missing or undefined values in one or more sensor axes was excluded prior to feature extraction [29]. Feature extraction and all related computations were implemented in Python (version 3.10), using NumPy, pandas, and SciPy. 3.3 Classification Model Development To classify cow behaviors, three Random Forest (RF) models were developed using features derived from accelerometer data, gyroscope data, and their combination. To preserve class distribution, the labeled dataset was initially split using stratified sampling into training/validation (80%) and testing (20%) subsets. The training/validation set was further divided (80:20) for hyperparameter tuning and model selection. Feature selection was conducted using Recursive Feature Elimination (RFE) with F1-score–based cross-validation, implemented through scikit-learn. RFE was applied independently for each cow-specific dataset to account for individual variability in movement patterns, consistent with individualized modeling practices in livestock behavior research [29]. The optimal number of features for each cow was determined by identifying the performance plateau in F1-score trends during Recursive Feature Elimination [31]. Model implementation was carried out in Python using the scikit-learn library, with support from pandas, NumPy, matplotlib, and joblib. The Random Forest (RF) algorithm, well suited for high-dimensional classification tasks, constructs an ensemble of decision trees, with final predictions determined by majority voting. Hyperparameters including the number of trees (n_estimators), maximum tree depth (max_depth), and minimum sample thresholds for node splitting (min_samples_split, min_samples_leaf) were optimized using grid search with 5-fold cross-validation. This approach enabled performance evaluation across rotating validation sets while mitigating overfitting. To ensure robustness, each model was evaluated over 10 independent iterations, and averaged performance metrics were reported. A fixed random seed was applied to each iteration to ensure reproducibility throughout the modeling process. Model performance was evaluated using standard multiclass classification metrics: accuracy, precision, sensitivity (recall), and F1-score. Each metric was calculated on a per-class basis using a one-vs-rest strategy and summarized through macro-averaging ensuring equal weighting across all behavior classes [32]. The following formulas were used to compute these metrics: Precision = TP / (TP + FP) Sensitivity (Recall) = TP / (TP + FN) F1-Score = 2 × (Precision × Recall) / (Precision + Recall) Accuracy = Correct predictions / Total predictions Where TP = true positives, FP = false positives, and FN = false negatives. To address class imbalance across behavior categories, the class_weight='balanced' parameter was performed during optimum feature selection and model training. Feature scaling was applied using StandardScaler within a pipeline structure to ensure consistency and compatibility with potential scale-sensitive model extensions. All modeling procedures were executed independently for each individual cow (Cow 1 to Cow 7) using a standardized and reproducible pipeline. 3.4 Statistical Analysis To preserve animal-level variability and avoid bias from data pooling, all analyses were conducted at the individual cow level (i.e., on a cow-by-cow basis). This approach enabled evaluation of both signal characteristics and machine learning (ML) performance within each animal, rather than aggregating results across the population. Accordingly, statistical comparisons and performance metrics were stratified by individual to capture between-animal heterogeneity in sensor signal patterns and behavioral expression. Statistical analyses were performed to evaluate differences in raw sensor signals and machine learning (ML) performance across four behavioral categories: lying, standing, eating, and walking. The Shapiro–Wilk test was used to assess normality and guide the selection of appropriate statistical tests. For normally distributed variables, one-way ANOVA followed by Tukey’s Honest Significant Difference (HSD) test was applied. When the assumption of normality was violated, the Kruskal–Wallis test with Dunn’s post hoc comparisons (Bonferroni-adjusted) was used [33, 34]. Homogeneity of variance was assumed for ANOVA and assessed through visual inspection of distributional characteristics. Statistical significance was defined as p < 0.05. Group-level differences were annotated using letter-based significance groupings generated by the multcompLetters and cldList functions. Comparisons were conducted across sensor types, signal axes (AccX, AccY, AccZ, GyroX, GyroY, GyroZ), as well as derived features such as signal magnitude area (SMA) and signal vector magnitude (SVM). Descriptive statistics including mean, standard deviation, range, median, and interquartile range were computed using R (version 4.3.1, RStudio Build 2023.06.1+524) with the dplyr, data.table, and rcompanion packages. To ensure reliable inference, only groups with a minimum of three valid observations were included in the analysis. Machine learning (ML) performance metrics, including accuracy, precision, sensitivity, and F1-score, were compared across sensor types for each behavior using the same test-selection strategy described above. This stratified approach accounted for differences in classification difficulty among behaviors [20]. Axis-specific comparisons within each cow and behavior were conducted using ANOVA followed by Tukey’s HSD test, with significant group differences indicated by letter-based annotations [28, 30]. All descriptive summaries, statistical results, and significance groupings were consolidated and exported using the openxlsx package. Statistical analyses were implemented in R using the FSA, rcompanion, multcompView, and ggplot2 packages. Results Sample Description Table 3. Number of cow data and distribution of behavioral instances across individual animals and activity classes. Cow No. Count of Labeled Behaviors Total Lying Standing Eating Walking 1 45,051 33,269 6,648 5,950 90,918 2 100,830 99,042 16,450 6,908 223,230 3 49,762 49,162 6,224 7,768 112,916 4 95,752 76,764 21,640 6,125 200,281 5 18,872 23,625 2,356 2,528 47,381 6 40,178 31,559 9,936 2,589 84,262 7 11,004 11,810 1,428 1,300 25,542 Number of Instances 361,449 325,231 64,682 33,168 784,530 Instances (%) 46.07 41.46 8.24 4.23 100.00 A total of 1,061,891 observations were collected from seven dairy cows over 90-day period using accelerometer and gyroscope sensors. After excluding technical failures, missing values, and estrus-related behaviors (53,095 observations), 784,530 observations (83.06%) remained for analysis, representing 1,794.1 hours of synchronized sensor and video-annotated data. Five cows contributed 90.7% of the dataset, while the remaining two accounted for 9.3%. The distribution of behavioral observations across the four target classes including lying, eating, standing, and walking is presented in Table 3. The dataset also included overlapping behaviors, such as rumination, which occasionally occurred concurrently with other activities. Behavioral Signal Patterns Tri-axial accelerometer data revealed distinct signal profiles corresponding to different behavioral states in dairy cows. Lying behavior was associated with low and stable values across all axes, particularly on the AccY axis, and corresponded to minimal movement. Moderate signal fluctuations were observed during standing and eating, with eating showing slightly greater variation. Walking behavior occurred intermittently and did not consistently result in elevated signal magnitudes across all axes. These patterns, derived from continuous 24-hour recordings, are visualized in Figure 3, which displays annotated time-series plots across seven individual cows. A clear diurnal rhythm was evident in the temporal distribution of behaviors. Walking and eating occurred more frequently during daylight hours (09:00–17:00), while lying behavior predominated during nighttime, reflecting rest-activity cycles typical of dairy cattle. Despite general consistency in these trends, subtle individual differences in signal amplitude and behavior duration were observed across cows. Tri-axial gyroscope data described rotational signal patterns across behaviors in dairy cows. Angular velocity profiles varied across behaviors, with lying consistently associated with low values across all axes. Standing showed moderate gyroscopic variation, while eating exhibited broader and more irregular angular fluctuations, particularly along the GyroX and GyroY axes, where several high-amplitude peaks were observed. Walking appeared intermittently and did not consistently produce distinguishable increases in angular velocity. Among all axes, GyroY demonstrated the largest dynamic range, which indicates higher sensitivity to rotational activity. Figure 4 also illustrates inter-animal variability in signal amplitude, suggesting cow-specific movement profiles within the same behavior. The signal vector magnitude derived from accelerometer data (SVM_acc) showed varying patterns across behaviors over a 24-hour period (Figure 5). Lying behavior consistently produced low SVM_acc values. Eating behavior was associated with irregular and elevated peaks in the signal. Standing behavior had moderate SVM_acc values with variable signal patterns, occasionally overlapping both the lower levels typical of lying and higher values seen in eating. Walking occurred intermittently and was not associated with clearly distinguishable changes in SVM_acc. Overall, SVM_acc patterns were similar across cows, although Cow 3 showed relatively higher variation in signal amplitude compared to others. Gyroscope-based Signal Vector Magnitude (SVM_gyro) showed changes in signal profiles across behaviors and cows. Lying behavior was consistently associated with low SVM_gyro values across all individuals. Elevated SVM_gyro values were observed during eating periods, with some instances exceeding 60°/s. Walking occurred sporadically and did not consistently produce elevated SVM_gyro levels. SVM_gyro profiles were generally similar across cows, although Cows 3 and 4 showed relatively larger variations in signal values, while Cows 2 and 5 displayed lower SVM_gyro amplitudes. These patterns are illustrated in Figure 6, which presents 24-hour SVM_gyro time-series plots for each individual cow. Sensor Signal Distribution Table 4. Descriptive statistics of Signal Vector Magnitude from accelerometer (SVM_acc) and gyroscope (SVM_gyro) across individual cows behaviors. Cow Behavior N SVM_acc SVM_gyro Mean±SE Min-Max Mean±SE Min-Max Cow No.1 Lying 45,051 0.33±0.19 b 0.01-2.19 7.34±8.84 b 0.01-79.48 Standing 33,269 0.31±0.2 c 0-2.23 18.82±14.87 c 0.07-80.12 Eating 6,648 0.45±0.19 a 0.02-1.81 28.93±13.79 a 1.12-80.12 Walking 5,950 0.31±0.2 c 0.01-1.92 29.36±14.12 a 0.46-80.12 Cow No.2 Lying 100,830 0.37±0.2 b 0-1.78 8.84±12.23 b 0.01-153.53 Standing 99,042 0.31±0.21 c 0-2.6 21.13±17.59 c 0.02-353.92 Eating 16,450 0.55±0.22 a 0.02-2.72 33.93±21.15 a 0.04-371.51 Walking 6,908 0.3±0.19 d 0.01-1.98 36.1±21.63 d 0.19-313.33 Cow No.3 Lying 49,762 0.33±0.15 b 0.01-1.47 6.56±8.94 b 0.02-78.39 Standing 49,162 0.38±0.18 c 0.01-1.8 20.41±16.37 c 0.03-81.61 Eating 6,224 0.54±0.19 a 0.03-2 32.58±15.12 a 1.57-81.23 Walking 7,768 0.38±0.18 d 0.03-1.77 35.51±14.56 d 0.6-81.51 Cow No.4 Lying 95,752 0.29±0.16 b 0-1.95 4.99±6.41 b 0.01-135 Standing 76,764 0.31±0.18 c 0.01-2.09 22.5±20.11 c 0.08-135 Eating 21,640 0.48±0.18 a 0.01-2.21 28.51±18.7 a 0.76-135 Walking 6,125 0.36±0.2 d 0.03-2.11 35.22±25.98 d 0.63-135 Cow No.5 Lying 18,872 0.26±0.14 b 0.01-1.15 8.34±14.26 b 0.04-181.6 Standing 23,625 0.28±0.17 c 0-1.38 32.91±29.92 c 0.14-193.7 Eating 2,356 0.47±0.18 a 0.02-1.29 52.9±38.97 a 0.56-192.98 Walking 2,528 0.33±0.17 d 0.02-1.08 48.28±34.04 a 1.81-192.33 Cow No.6 Lying 40,178 0.28±0.16 b 0-1.76 5.27±6.2 b 0.05-52.69 Standing 31,559 0.31±0.18 c 0.01-1.95 21.71±15.13 c 0.22-52.69 Eating 9,936 0.48±0.17 a 0.04-2.15 27.25±13.52 a 0.76-52.69 Walking 2,589 0.35±0.2 d 0.03-2.01 30.9±14.64 d 0.63-52.69 Cow No.7 Lying 11,004 0.3±0.15 b 0.03-1.67 4.09±6.08 b 0.02-40.76 Standing 11,810 0.25±0.18 c 0-1.57 15.96±11.98 c 0.04-40.88 Eating 1,428 0.5±0.18 a 0.11-1.16 28.37±9.19 a 3.63-40.88 Walking 1,300 0.34±0.18 d 0.03-1.03 28.77±9.5 a 3.08-40.88 Different lowercase superscript letters (a–d) indicate statistically significant differences (p < 0.05) of mean values among behaviors. Descriptive analysis of Signal Vector Magnitude from accelerometer (SVM_acc) and gyroscope (SVM_gyro) data revealed clear behavioral differentiation across the four target activities. Lying behavior consistently produced the lowest mean values for both SVM_gyro and SVM_acc, reflecting minimal body movement and rotational activity, although for SVM_acc, this pattern was observed in five out of seven cows. In contrast, eating produced the highest average SVM values, driven by dynamic head and neck motion. Walking exhibited similar SVM_gyro levels to eating in several cows, indicating comparable rotational intensity during these behaviors. Standing, as expected, presented intermediate values. Statistical comparisons, conducted at the individual-animal level and reported in Table 4, confirmed significant differences between behaviors (p < 0.05), as denoted by differing lowercase subscripts. However, in Cows 1, 5, and 7, no significant difference was detected between eating and walking for SVM_gyro, suggesting overlapping motion patterns in these more active states. These findings emphasize both the discriminative power of SVM features and the presence of cow-specific variation in behavior-related movement dynamics. Table 5. Descriptive statistics of raw tri-axial accelerometer signals (AccX, AccY, AccZ) across behaviors in individual cows. Cow Behavior N AccX AccY AccZ Mean±SE Min-Max Mean±SE Min-Max Mean±SE Min-Max Cow No.1 Lying 45,051 0.13±0.28 a,A -1.4-2.16 -0.08±0.09 a,B -0.49-0.3 0.07±0.19 a,C -0.96-1.04 Standing 33,269 -0.04±0.29 b,A -1.94-1.79 -0.07±0.09 b,B -0.49-0.3 0±0.18 b,C -1.05-1.14 Eating 6,648 -0.3±0.26 d,A -1.42-1.67 -0.12±0.13 d,B -0.47-0.3 -0.11±0.21 d,B -1.06-1.14 Walking 5,950 -0.14±0.24 c,A -1.67-1.7 -0.06±0.11 c,B -0.49-0.3 -0.06±0.2 c,C -1.06-1.14 Cow No.2 Lying 100,830 0.14±0.21 a,A -0.83-1.7 -0.07±0.09 a,B -0.96-0.52 -0.16±0.28 a,C -1.16-1.2 Standing 99,042 -0.02±0.21 b,A -2.1-2.08 -0.06±0.09 b,B -1.31-1 -0.16±0.24 a,C -2-1.57 Eating 16,450 -0.31±0.25 d,A -2-1.69 -0.15±0.12 d,B -1.31-0.5 -0.31±0.25 c,C -2-0.78 Walking 6,908 -0.08±0.2 c,A -1.61-1.42 -0.05±0.11 c,B -1.31-1 -0.1±0.23 b,A -1.52-1.54 Cow No.3 Lying 49,762 0.09±0.19 a,A -0.7-1.23 -0.05±0.06 a,B -0.38-0.28 -0.19±0.2 a,C -1.39-0.92 Standing 49,162 -0.01±0.28 b,A -1.39-1.47 -0.07±0.1 b,B -0.38-0.28 -0.24±0.17 b,C -1.39-0.92 Eating 6,224 -0.29±0.26 d,A -1.39-1.43 -0.13±0.13 c,B -0.38-0.28 -0.34±0.18 d,C -1.39-0.91 Walking 7,768 -0.04±0.24 c,A -1.39-1.47 -0.06±0.12 a,B -0.38-0.28 -0.27±0.17 c,C -1.38-0.81 Cow No.4 Lying 95,752 0.12±0.22 a,A -1.67-1.94 -0.06±0.07 a,B -0.38-0.28 -0.07±0.18 a,C -1.22-0.98 Standing 76,764 -0.03±0.26 b,A -1.91-1.99 -0.06±0.08 b,B -0.38-0.28 -0.17±0.15 b,C -1.22-0.98 Eating 21,640 -0.25±0.26 d,A -1.97-1.99 -0.11±0.1 c,B -0.38-0.28 -0.28±0.17 d,C -1.22-0.98 Walking 6,125 -0.07±0.28 c,A -1.63-1.94 -0.06±0.1 a,B -0.38-0.28 -0.21±0.17 c,C -1.22-0.98 Cow No.5 Lying 18,872 0.11±0.2 a,A -1.14-0.98 -0.05±0.05 a,B -0.32-0.21 -0.08±0.14 a,C -0.71-0.55 Standing 23,625 -0.08±0.22 b,A -1.37-0.97 -0.06±0.07 b,B -0.32-0.23 -0.16±0.14 b,C -0.92-0.55 Eating 2,356 -0.28±0.24 d,A -1.13-0.84 -0.11±0.1 d,B -0.32-0.23 -0.26±0.16 d,C -0.92-0.55 Walking 2,528 -0.16±0.21 c,A -0.9-0.72 -0.06±0.09 c,B -0.32-0.23 -0.18±0.16 c,C -0.92-0.55 Cow No.6 Lying 40,178 0.12±0.22 a,A -1.67-1.74 -0.06±0.07 a,B -0.32-0.23 -0.06±0.17 a,C -1.22-0.98 Standing 31,559 -0.04±0.25 b,A -1.67-1.74 -0.06±0.08 b,B -0.32-0.23 -0.18±0.15 b,C -1.22-0.98 Eating 9,936 -0.27±0.23 d,A -1.67-1.74 -0.11±0.1 c,B -0.32-0.23 -0.3±0.16 d,C -1.22-0.98 Walking 2,589 -0.07±0.26 c,A -1.63-1.74 -0.06±0.1 b,B -0.32-0.23 -0.22±0.16 c,C -1.22-0.98 Cow No.7 Lying 11,004 0.04±0.13 a,A -0.58-1 -0.05±0.06 a,B -0.31-0.24 -0.23±0.18 a,C -1.39-0.91 Standing 11,810 -0.05±0.23 b,A -0.98-1 -0.05±0.08 b,B -0.31-0.24 -0.09±0.16 b,C -1.48-1.16 Eating 1,428 -0.34±0.19 d,A -0.98-0.63 -0.14±0.11 d,B -0.31-0.24 -0.24±0.21 a,C -0.98-0.59 Walking 1,300 -0.22±0.22 c,A -0.98-0.44 -0.07±0.12 c,B -0.31-0.24 -0.14±0.14 c,C -0.61-0.49 Different lowercase superscript letters (a–d) indicate statistically significant differences of raw tri-axial accelerometer signal among behaviors (p < 0.05). Uppercase superscript letters (A-C) indicate significant differences of raw tri-axial accelerometer signal among axes (X, Y, and Z) within each behavior (p < 0.05). Analysis of raw tri-axial accelerometer signals (AccX, AccY, AccZ) revealed distinct patterns across behavioral states, as detailed in Table 5. Lying and standing exhibited the lowest signal variation across all three axes, consistent with their classification as low-movement or static behaviors. In contrast, eating behavior was characterized by negative mean values and high standard deviations, particularly in AccX and AccY, indicating frequent and variable head and neck movements. Walking also showed elevated variability, though generally less pronounced than eating. Table 6. Descriptive statistics of raw tri-axial gyroscope signals (GyroX, GyroY, GyroZ) across behaviors in individual cows. Cow Behavior N GyrX GyrY GyrZ Mean±SE Min-Max Mean±SE Min-Max Mean±SE Min-Max Cow No.1 Lying 45,051 -0.49±4.32 a,A -37.34-33.35 0.68±7.77 a,B -63.84-62.88 -1.27±7.13 a,C -31.41-33.6 Standing 33,269 -0.27±10.72 b,A -37.34-33.35 0.89±18.65 a,B -63.84-62.88 -0.67±10.55 b,C -31.41-33.6 Eating 6,648 0.33±17.16 c,A -37.34-33.35 1.75±22.92 b,B -63.84-62.88 -0.86±14.27 b,C -31.41-33.6 Walking 5,950 -0.22±15.62 ab,A -37.34-33.35 0.99±24.75 a,B -63.84-62.88 -0.75±14.27 b,A -31.41-33.6 Cow No.2 Lying 100,830 0.79±4.5 a,A -36.54-52.89 0.08±13.86 a,B -147.92-151.69 0.21±3.85 a,C -67.61-67.03 Standing 99,042 1.07±10.12 b,A -36.54-52.89 -1.16±23.61 b,B -207.18-350.75 0.11±9.68 b,C -67.61-67.03 Eating 16,450 1.98±16.56 c,A -36.54-52.89 -3.45±33.39 c,B -361.56-350.75 -0.03±13.92 b,C -67.61-67.03 Walking 6,908 1.83±16.45 c,A -36.54-52.89 -2.89±35.68 c,B -259.52-307.69 0.42±14.69 a,C -67.61-67.03 Cow No.3 Lying 49,762 -0.19±4.97 a,A -39.8-49.44 -0.26±8.54 a,B -47.4-48.96 0.07±5.02 a,AB -44.36-49.2 Standing 49,162 -0.16±11.04 a,A -39.8-49.44 0.27±20.51 b,B -47.4-48.96 0.69±11.88 b,B -44.36-49.2 Eating 6,224 0.29±16.3 b,A -39.8-49.44 3.89±27.14 d,B -47.4-48.96 -0.71±16.5 d,C -44.36-49.2 Walking 7,768 -0.36±16.2 a,A -39.8-49.44 -1.97±29.96 c,B -47.4-48.96 2.25±17.44 c,C -44.36-49.2 Cow No.4 Lying 95,752 1.17±3.69 a,A -58.24-57.92 2.68±6.03 a,B -112.12-114.15 -0.15±2.73 a,C -42.48-41.78 Standing 76,764 1.2±13.38 ab,A -58.24-57.92 1.99±24.79 b,B -112.12-114.15 -0.01±10.57 b,C -42.48-41.78 Eating 21,640 1.41±17.55 b,A -58.24-57.92 2.23±26.78 b,B -112.12-114.15 -0.1±11.41 ab,C -42.48-41.78 Walking 6,125 0.37±19.77 c,A -58.24-57.92 0.37±36.21 c,B -112.12-114.15 0.66±14.59 c,C -42.48-41.78 Cow No.5 Lying 18,872 -0.39±6.6 a,A -73.1-71.17 1.31±13.7 a,B -152.39-166.64 0±6.3 a,C -62.29-66.41 Standing 23,625 -0.69±16.29 a,A -73.1-71.17 4.51±37.75 b,B -152.39-166.64 -1.05±16.32 b,A -62.29-66.41 Eating 2,356 1.82±26.65 c,A -73.1-71.17 -0.91±55.06 d,B -152.39-166.64 4.61±23.48 c,C -62.29-66.41 Walking 2,528 -2.4±23.31 b,A -73.1-71.17 7.68±48.46 c,B -152.39-166.64 -1.68±23.04 b,A -62.29-66.41 Cow No.6 Lying 40,178 1.35±3.67 ab,A -24.08-26.52 2.98±5.95 a,B -36.94-42.37 -0.15±2.58 a,C -16.65-16.46 Standing 31,559 1.29±12.4 a,A -24.08-26.52 2.53±21.33 b,B -36.94-42.37 -0.09±9.13 ab,C -16.65-16.46 Eating 9,936 1.61±16.19 b,A -24.08-26.52 2.55±23.47 ab,B -36.94-42.37 0.07±10.17 b,C -16.65-16.46 Walking 2,589 0.58±16.53 c,A -24.08-26.52 0.85±27.61 c,B -36.94-42.37 0.52±11.5 c,C -16.65-16.46 Cow No.7 Lying 11,004 0.04±3.58 a,A -20.09-19.84 0.04±5.81 a,B -30.79-31.91 0.28±2.64 ab,A -14.78-15.8 Standing 11,810 -0.62±8.91 b,A -20.09-19.84 1.87±15.77 b,B -30.79-31.91 1.31±8.03 c,B -14.78-15.8 Eating 1,428 -0.81±14.38 abc,A -20.09-19.84 4.99±23.26 c,B -30.79-31.91 -0.19±10.86 b,A -14.78-15.8 Walking 1,300 -1.74±15.43 c,A -20.09-19.84 2.84±23.15 bc,B -30.79-31.91 1.09±11.59 ac,AB -14.78-15.8 Different lowercase superscript letters (a–d) indicate statistically significant differences of raw tri-axial gyroscope signal among behaviors (p < 0.05). Uppercase superscript letters (A-C) indicate significant differences of raw tri-axial gyroscope signal among axes (X, Y, and Z) within each behavior (p < 0.05). Tri-axial gyroscope signals (GyroX, GyroY, GyroZ) demonstrated clear behavioral distinctions in rotational movement patterns, as detailed in Table 6. Lying and standing were consistently associated with the lowest mean values and minimal variability across all axes, reflecting limited angular motion during these postural states. In contrast, eating and walking produced markedly higher gyroscopic activity, particularly along the GyroY and GyroZ axes, indicative of frequent head and body rotations. Model Performance Table 7. Classification performance of Random Forest models using accelerometer, gyroscope, and combined features across four behaviors. Behavior Device Precision Sensitivity F1-score Overall Accuracy N Mean ± SE Min-Max Mean ± SE Min-Max Mean ± SE Min-Max Mean ± SE Min-Max Lying Acc 70 93.33±2.45 ab 90.75-96.87 94.52±1.6 a 92.01-96.51 93.91±1.93 a 91.55-96.69 89.44±3.5 a 86.26-94.15 Gyr 70 92.93±2.07 b 90.39-96.87 94.5±1.91 a 91.9-96.94 93.71±1.88 a 91.14-96.91 88.35±3.63 a 83.46-93.56 Com 70 95.94±1.66 a 94.02-98.21 96.81±1.25 b 94.8-98.37 96.37±1.34 b 94.81-98.29 92.74±2.68 b 89.99-96.77 Standing Acc 70 85.44±4.99 a 80.04-92.71 90.28±4.33 a 84.51-95.56 87.78±4.57 a 82.68-93.93 89.44±3.5 a 86.26-94.15 Gyr 70 83.63±5.66 a 75.52-92.2 90.81±2.81 a 87.76-95.87 87.03±4.11 a 81.33-92.44 88.35±3.63 a 83.46-93.56 Com 70 89.63±3.9 a 84.4-95.91 93.67±3.02 a 89.72-97.61 91.59±3.34 a 88.21-96.75 92.74±2.68 b 89.99-96.77 Eating Acc 70 83.61±5.93 a 77.67-94.17 65.65±10.75 a 54.05-84.54 73.29±8.3 a 63.75-89.1 89.44±3.5 a 86.26-94.15 Gyr 70 82.65±7.83 a 74.57-96.15 42.44±13.78 b 30.23-71.92 55.4±12.59 b 44.83-82.29 88.35±3.63 a 83.46-93.56 Com 70 86.29±4.53 a 83.64-95.95 72.46±8.44 a 62.16-87.32 78.65±6.57 a 71.32-91.44 92.74±2.68 b 89.99-96.77 Walking Acc 70 90.65±7.35 a 82.06-100 31.78±17.23 a 5.34-59.18 44.65±20.3 a 10.15-74.22 89.44±3.5 a 86.26-94.15 Gyr 70 91.33±9.77 a 75.56-99.76 34.34±16.8 a 15-62.76 47.62±17.33 a 26.01-77.05 88.35±3.63 a 83.46-93.56 Com 70 93.26±5.96 a 83.64-100 46.76±16.22 a 20.34-65.52 60.61±15.8 a 33.67-78.89 92.74±2.68 b 89.99-96.77 Different lowercase superscript letters (a–d) indicate significant differences (p < 0.05) of mean classification performance among device types (accelerometer, gyroscope, combination), based on 70 data points (10 iterations per cow × 7 cows). Bolded values represent the highest means for each performance metric across device types (accelerometer, gyroscope, and combination) within each behavioral classification. Classification performance metrics including precision, sensitivity, F1-score, and overall accuracy varied across behaviors and sensor configurations, as summarized in Table 7. For lying, the combined-sensor model significantly outperformed both single-sensor models across all metrics (p < 0.05), demonstrating its advantage in detecting low-movement behaviors. In the case of standing, overall accuracy was significantly higher for the combined model, although differences in precision, sensitivity, and F1-score were not statistically significant. For eating, the combined model yielded significantly higher sensitivity and F1-score compared to the gyroscope-only model (p < 0.05), while precision did not differ significantly across configurations. Walking proved to be the most challenging behavior to classify, with all models showing relatively low sensitivity and F1-score. However, the combined model achieved significantly higher overall accuracy for this class, suggesting modest gains through sensor fusion even in more dynamic and variable behaviors. Individual-level results, presented in Supplementary Table 1, revealed notable inter-cow variability. Nevertheless, the combined model generally produced higher or more consistent performance across cows, particularly for lying and standing. Classification performance for eating and walking was more variable, both across sensor types and among individual cows, reflecting the complexity and heterogeneity of these behaviors. Discussion This study aimed to examine the characteristics of accelerometer and gyroscope signals across four behavioral categories (lying, standing, eating, and walking) and to evaluate the effect of different sensor configurations on classification performance in individual dairy cows. The analysis revealed clear behavioral distinctions in the sensor signal patterns. For instance, lying behavior was associated with lower signal variability, particularly in the gyroscope Z-axis and accelerometer Z-axis. In contrast, walking consistently exhibited higher signal intensity, especially in the accelerometer X-axis and gyroscope X- and Y-axes. Eating, however, was characterized by elevated and irregular signal amplitudes across both sensor types and multiple axes, notably in GyroY and AccX, reflecting complex head and neck movements. These findings underscore the potential of axis-specific features in distinguishing movement states. The 10-second window length used for segmentation was selected to balance temporal granularity with behavioral stability, allowing for representative movement capture while minimizing noise. Regarding classification performance, models utilizing both accelerometer and gyroscope data achieved the highest accuracy and F1-scores across all behaviors, outperforming those using accelerometer-only or gyroscope-only inputs. This advantage was particularly pronounced in behaviors such as lying and standing, which exhibited more subtle signal dynamics. Nevertheless, performance varied across individual cows and behaviors, particularly in eating and walking, reflecting the influence of cow-specific movement patterns and the inherent complexity of differentiating between dynamic behaviors. These results collectively highlight the importance of incorporating both translational and rotational motion data to enhance behavioral recognition in dairy cows. 1. Behavior-Specific Sensor Characteristics Across Different Axes The tri-axial analysis of accelerometer and gyroscope signals revealed distinct axis-specific sensitivities associated with different behavioral states. For accelerometer data, the Z-axis consistently exhibited low and stable values during lying, reflecting minimal vertical displacement, while eating behavior produced the largest deviations, especially along the X- and Y-axes, indicating active head and neck movement. Standing behavior exhibited intermediate signal values across all axes, while walking did not consistently produce elevated or distinctive signal magnitudes, suggesting it may be less distinguishable using raw accelerometer data alone. These findings align with previous studies indicating that posture-related behaviors typically generate clearer accelerometric patterns than locomotor activities [9, 10]. For gyroscope signals, all three axes remained relatively stable during lying, with GyroZ showing the lowest variation across cows. In contrast, eating was characterized by the highest rotational intensity, particularly in the GyroY and GyroZ axes, suggesting active and frequent head rotations during this behavior. Standing behavior exhibited moderate angular variation, while walking produced variable, cow-dependent signal patterns that occasionally overlapped with eating in both amplitude and distribution. This is consistent with previous gyroscope-based studies, which have shown that rotational movement patterns are more effective for detecting feeding and transition behaviors than for identifying consistent walking signatures [30]. Notably, signal variability differed among cows even within the same behavior, highlighting inter-individual differences in movement patterns. These may reflect differences in body conformation, locomotion patterns, or environmental conditions such as pen structure or feeding space. Such inter-animal variation has been noted in earlier sensor-based studies and underscores the value of personalized models in behavior monitoring [9, 35]. These cow-specific patterns were evident in both signal amplitude and axis distribution, reinforcing the importance of individualized analysis in behavior classification systems. Overall, these findings underscore the importance of directional sensitivity in interpreting behavioral signals. Axis-specific features, particularly from GyroY and AccX/AccY during active behaviors, offer valuable discriminative power and should be prioritized in sensor placement strategies and feature selection processes for future precision livestock monitoring applications [28, 36, 37]. 2. Classification Performance and Model Behavior The integration of accelerometer and gyroscope data consistently produced superior classification performance across behaviors and individual cows, highlighting the advantages of sensor fusion for behavioral modeling. This enhancement is likely attributable to the complementary nature of translational and rotational data streams, wherein accelerometers effectively capture posture-related changes, and gyroscopes provide critical information on dynamic orientation, particularly during locomotor activities. Similar improvements in classification performance achieved through sensor fusion have been reported in recent precision livestock studies [17, 38]. Across all behaviors, the accelerometer-alone model generally outperformed the gyroscope-alone model, especially for static behaviors such as lying and standing. This trend suggests that accelerometer-derived features provide more distinguishable signals for postural states, whereas gyroscope signals tend to exhibit greater overlap between similar behaviors, a challenge also noted in livestock monitoring studies using ensemble classifiers [20]. Among the four behaviors, walking, which involves pronounced dynamic movement, posed the greatest classification challenge, with all sensor configurations demonstrating lower F1-scores and higher inter-cow variability. Despite these challenges, the combined-sensor model demonstrated improved accuracy and robustness, particularly notable in cows with low sensitivity in the gyroscope-only model. It is important to note that behaviors such as eating and walking were relatively underrepresented in the dataset. To mitigate potential classification bias arising from this class imbalance, weighted class penalties were applied during model training. This approach ensured that less frequent behaviors contributed proportionately to the learning process, thereby enhancing the model’s sensitivity to underrepresented behaviors. These findings underscore the capacity of Random Forest classifiers to exploit multidimensional, nonlinear features and effectively integrate heterogeneous sensor modalities for reliable behavior recognition across diverse individuals. The algorithm’s ensemble-based architecture enables it to capture nuanced, behavior-specific patterns within a high-dimensional feature space, reinforcing its suitability for deployment in precision livestock monitoring systems [39]. 3. Contextualization within Prior Literature The findings of this study are broadly consistent with prior research that has employed accelerometer-only or gyroscope-only data for classifying cattle behavior, yet they extend the literature by offering a more nuanced axis-specific and individual-level analysis. Previous studies using accelerometer data alone have demonstrated acceptable classification performance for postural behaviors, particularly for lying and standing, due to the stability and directionality of gravitational acceleration [10]. Similarly, gyroscope-based models have shown advantages in capturing the rotational aspects of movement, especially during walking or transitional movements [17]. The present study advances beyond previous sensor-level approaches by decomposing axis-specific signal patterns and illustrating their behavioral relevance in a structured, cow-specific context. Unlike many earlier studies that aggregated data across animals, this investigation employed modeling at the individual-cow level, revealing behavior-specific variability in signal characteristics and classification performance, particularly in eating and walking. These findings suggest that cow-specific locomotor patterns may influence model performance, an aspect often underexplored in pooled analyses. Moreover, the comparative evaluation of sensor fusion demonstrated additive benefits, confirming that combining accelerometer and gyroscope data enhances model robustness across behavior types consistent with more recent multi-sensor livestock monitoring studies [16, 36]. Some divergences from earlier findings may be attributable to differences in sensor placement, window length, sampling rate, or behavioral labeling strategies. 4. Strengths, Limitations, Application, and Future Directions This study presents several methodological strengths that enhance its scientific rigor and practical relevance. By conducting behavioral classification at the level of individual cows, the analysis captures inter-animal variability that is often overlooked in pooled modeling approaches. This individualized framework allows for a more precise understanding of cow-specific movement signatures and behavior expressions. Additionally, the structured axis-level evaluation of both accelerometer and gyroscope signals provides granular insight into the contribution of each sensor dimension, advancing the field’s understanding of movement ecology in dairy cattle. The use of Random Forest, a highly interpretable machine learning algorithm, further supports transparency and practical applicability through feature importance ranking, which aids in refining sensor deployment strategies and model development pipelines. Random Forest was selected due to its robustness to overfitting, capacity to manage high-dimensional and imbalanced datasets, and minimal assumptions about data distribution [40]. Its built-in feature importance metrics also support model interpretation, which is especially useful for sensor signal classification in animal behavior studies [20, 41]. Nevertheless, certain limitations should be acknowledged. The relatively small sample size which was limited to seven heifers housed under consistent conditions may constrain the generalizability of findings to other dairy populations or production systems. However, the depth, duration, and resolution of the data collected (90 days of continuous sensing and over 780,000 labeled observations) provide strong analytical power for within-cow modeling. Although behavior annotation was performed manually, it followed a robust dual-observer protocol with high agreement (κ = 0.84), supported by joint training and consensus resolution. This enhances label reliability, even though manual annotation is time-consuming and may limit scalability in larger herds. The scope of labeled behaviors was restricted to four clearly defined categories including lying, standing, eating, and walking while others such as ruminating or drinking were excluded due to low frequency or substantial behavioral overlap. This trade-off improved classification clarity but may limit behavioral comprehensiveness. Sensor placement was limited to the neck, which aligns with commercial practices but may reduce sensitivity to limb movement or asymmetric postures. An additional limitation involves the unequal distribution of behavioral data across individual cows, with some contributing disproportionately to specific behaviors such as walking and eating. This imbalance may introduce bias in model training and affect generalizability across individuals. However, this study addressed the issue through class-weighted modeling to ensure the representation of less frequent behaviors and by implementing cow-level analysis to preserve inter-individual variability. Class-weighted training has been validated as an effective strategy for handling imbalanced livestock behavior data, improving recognition of minority classes [42]. Furthermore, individual-cow modeling has been emphasized as a crucial technique for capturing heterogeneous behavioral expression in dairy cattle, which pooled models often overlook [13, 43]. These findings support the use of individualized modeling pipelines for precision livestock systems, particularly when behavior frequency and movement patterns differ markedly between animals. For practical implementation in commercial dairy systems, individualized behavior classification models must be supported by scalable calibration and rigorous validation under real-world conditions. The proposed models can be embedded in edge computing platforms integrated with wearable sensor units, enabling real-time feature extraction and behavior classification directly on the animal. This setup reduces reliance on continuous wireless transmission and supports low-latency monitoring. Data aggregation and advanced analytics can be conducted on centralized farm servers or cloud platforms, where long-term trends in individual behavior can inform herd-level decision-making. To ensure adaptability to animal-specific behavior patterns, which is critical for the early detection of health and welfare deviations, future applications should incorporate flexible calibration workflows. These workflows could involve brief training periods using labeled behavioral data, such as short-term video annotations, to personalize models for newly introduced animals or production environments. Additionally, user interfaces should be developed to integrate model outputs into existing farm management software, thereby enabling automated alerts and decision-support tools for health surveillance, nutritional adjustments, and welfare monitoring. In addition to calibration, comprehensive external validation of the best-performing models across diverse environmental and management conditions is essential. This requirement is particularly relevant in regions with varied environmental and management conditions, such as tropical areas where dairy operations often include loose housing, tie-stall barns, and pasture-based systems. Each of these settings presents distinct challenges related to behavioral expression, sensor signal quality, and variability in daily routines. Validation in such contexts is necessary not only to evaluate model robustness and adaptability, but also to refine behavioral thresholds, calibration strategies, and sensor configurations suited to heterogeneous farm environments. Rather than serving solely as a final performance check, validation should be considered an integral part of the model development process, contributing to the generalizability, operational reliability, and practical deployment of individualized behavior monitoring systems in precision livestock farming. Future research should aim to build on these findings by incorporating larger and more heterogeneous cow populations across multiple farms and management systems. Validation under commercial conditions will be essential to ensure the robustness and scalability of the proposed models. Furthermore, the integration of deep learning and architectures, particularly those designed for temporal or sequential data (e.g., convolutional or recurrent neural networks), may enhance classification accuracy by learning context-aware movement patterns across time windows [44-46]. Finally, extending behavioral coverage to include overlapping or complex behaviors, such as ruminating, drinking, or social interaction, would increase the ecological relevance of automated monitoring systems and support broader applications in precision livestock management. Conclusion This study demonstrated that integrating accelerometer and gyroscope data improves the accuracy and robustness of dairy cow behavior classification, particularly for static behaviors such as lying and standing. Axis-specific analysis revealed that both translational and rotational features contributed significantly to behavior differentiation, while individual-level modeling captured cow-specific variation in movement patterns. The Random Forest classifier performed reliably across behaviors and individuals by utilizing high-dimensional, multimodal features. Despite limitations related to sample size and herd homogeneity, this work establishes a strong foundation for sensor-based behavior monitoring in precision livestock systems. Future research should aim to validate these findings across diverse herds and environments, explore temporal deep learning architectures, and expand behavior recognition to include overlapping or complex activities. Abbreviations AccX, AccY, AccZ Accelerometer signals along X, Y, and Z axes ANOVA Analysis of Variance CCTV Closed-Circuit Television DBA Dynamic Body Acceleration DBAX, DBAY, DBAZ Dynamic Body Acceleration on X, Y, and Z axes FN False Negative FP False Positive GyroX, GyroY, GyroZ Gyroscope signals along X, Y, and Z axes HSD Honest Significant Difference IQR Interquartile Range ML Machine Learning ODBA Overall Dynamic Body Acceleration RFE Recursive Feature Elimination RF Random Forest RMS Root Mean Square SE Standard Error SMA Signal Magnitude Area SVM Signal Vector Magnitude SVM_acc Signal Vector Magnitude from accelerometer SVM_gyro Signal Vector Magnitude from gyroscope TP True Positive Wi-Fi Wireless Fidelity VeDBA Vectorial Dynamic Body Acceleration Declarations · Ethics approval and consent to participate This study was approved by the Institutional Animal Care and Use Committee (IACUC) of the Faculty of Veterinary Science, Chulalongkorn University, Thailand (Protocol No. 2031047), in accordance with institutional regulations and the Ethical Principles and Guidelines for the Use of Animals for Scientific Purposes issued by the National Research Council of Thailand. · Consent for publication Not applicable · Availability of data and materials The cow activity data generated and analyzed during this study is available upon reasonable request from the Research Unit of Data Innovation for Livestock, Department of Veterinary Medicine, Faculty of Veterinary Science, Chulalongkorn University, Bangkok, Thailand. To support reproducibility, a subset of the data comprising both raw and processed activity data is publicly available at the GitHub repository: https://github.com/pongsanun. In addition, all Python scripts used for data manipulation, feature engineering, visualization, and model development, as well as R scripts used for statistical analysis, are provided in the same repository. · Competing interests The authors declare that they have no competing interests · Funding This research was supported by the Research and Researchers for Industries (RRI) program (Contract No. PHD60I0084), the Program Management Unit for Competitiveness (PMUC) (Contract No. 1499287), and the Research Unit of Data Innovation for Livestock, Department of Veterinary Medicine, Faculty of Veterinary Science, Chulalongkorn University, Bangkok, Thailand. · Authors' contributions PK and CI were responsible for the study design and overall conceptualization. AT, AL, and SC contributed to the design and development of the sensor prototypes and data transmission systems used for data collection. PK was responsible for prototype installation and system monitoring to ensure successful data acquisition. Behavioral data annotation was conducted by PK and AT. All data analyses were performed by PK. The manuscript was drafted by PK with critical revisions and editorial input from CI. All authors have read and approved the final version of the manuscript. · Acknowledgements The authors would like to express their sincere gratitude to the Director of the Farm Animal Hospital, Faculty of Veterinary Science, Chulalongkorn University, Nakhon Pathom, Thailand, for granting permission to conduct the experiment. The authors also extend their heartfelt appreciation to the animal husbandry team and hospital staff for their continuous support and kind assistance throughout the experimental period. · Authors' information (optional) References Qiao Y, Guo Y, Yu K, He D: C3D-ConvLSTM based cow behaviour classification using video data for precision livestock farming. Computers and Electronics in Agriculture 2022, 193:106650. Norton T, Berckmans D: Developing precision livestock farming tools for precision dairy farming. Animal Frontiers 2017, 7(1):18-23. Halachmi I, Guarino M: Editorial: Precision livestock farming: a ‘per animal’ approach using advanced monitoring technologies. animal 2016, 10(9):1482-1483. Walker SL, Smith RF, Routly JE, Jones DN, Morris MJ, Dobson H: Lameness, Activity Time-Budgets, and Estrus Expression in Dairy Cattle. Journal of Dairy Science 2008, 91(12):4552-4559. Weigele HC, Gygax L, Steiner A, Wechsler B, Burla JB: Moderate lameness leads to marked behavioral changes in dairy cows. Journal of Dairy Science 2018, 101(3):2370-2382. Pereira GM, Sharpe KT, Heins BJ: Evaluation of the RumiWatch system as a benchmark to monitor feeding and locomotion behaviors of grazing dairy cows. Journal of Dairy Science 2021, 104(3):3736-3750. Riaboff L, Shalloo L, Smeaton AF, Couvreur S, Madouasse A, Keane MT: Predicting livestock behaviour using accelerometers: A systematic review of processing techniques for ruminant behaviour prediction from raw accelerometer data. Computers and Electronics in Agriculture 2022, 192:106610. Benaissa S, Tuyttens FAM, Plets D, Martens L, Vandaele L, Joseph W, Sonck B: Improved cattle behaviour monitoring by combining Ultra-Wideband location and accelerometer data. animal 2023, 17(4):100730. Robért B, White B, Renter D, Larson R: Evaluation of three-dimensional accelerometers to monitor and classify behavior patterns in cattle. Computers and Electronics in Agriculture 2009, 67:80-84. Martiskainen P, Järvinen M, Skön J-P, Tiirikainen J, Kolehmainen M, Mononen J: Cow behaviour pattern recognition using a three-dimensional accelerometer and support vector machines. Applied Animal Behaviour Science 2009, 119(1):32-38. Pereira GM, Heins BJ, Endres MI: Technical note: Validation of an ear-tag accelerometer sensor to determine rumination, eating, and activity behaviors of grazing dairy cattle. Journal of Dairy Science 2018, 101(3):2492-2495. Russel NS, Selvaraj A: Decoding cow behavior patterns from accelerometer data using deep learning. Journal of Veterinary Behavior 2024, 74:68-78. Williams ML, Mac Parthaláin N, Brewer P, James WPJ, Rose MT: A novel behavioral model of the pasture-based dairy cow from GPS data using data mining and machine learning techniques. Journal of Dairy Science 2016, 99(3):2063-2075. Anzai H, Hirata M: Individual Monitoring of Behavior to Enhance Productivity and Welfare of Animals in Small-Scale Intensive Cattle Grazing Systems. Frontiers in Sustainable Food Systems 2021, Volume 5 - 2021. Singhal G, Choudhary P, Abhishek V, Sweety S, Subramanian S, Goel N: Cattle Collar: An End-to-End Multi-Model Framework for Cattle Monitoring. In: 2022 IEEE 5th International Conference on Multimedia Information Processing and Retrieval (MIPR): 2-4 Aug. 2022 2022; 2022: 401-407. Benaissa S, Tuyttens FAM, Plets D, de Pessemier T, Trogh J, Tanghe E, Martens L, Vandaele L, Van Nuffel A, Joseph W et al: On the use of on-cow accelerometers for the classification of behaviours in dairy barns. Research in Veterinary Science 2019, 125:425-433. Mladenova T, Valova I, Evstatiev B, Valov N, Varlyakov I, Markov T, Stoycheva S, Mondeshka L, Markov N: Evaluation of the Efficiency of Machine Learning Algorithms for Identification of Cattle Behavior Using Accelerometer and Gyroscope Data. In: AgriEngineering. vol. 6; 2024: 2179-2197. Obermeyer K, Kayser M: On-farm assessment of grazing behaviour of dairy cows in two pasture management systems by low-cost and reliable cowtrackers. Smart Agricultural Technology 2023, 6:100349. Williams M, Zhan Lai S: Classification of dairy cow excretory events using a tail-mounted accelerometer. Computers and Electronics in Agriculture 2022, 199:107187. Martono NP, Sawado R, Nonaka I, Terada F, Ohwada H: Automated Cattle Behavior Classification Using Wearable Sensors and Machine Learning Approach. In: Knowledge Management and Acquisition for Intelligent Systems: 2023// 2023; Singapore: Springer Nature Singapore; 2023: 58-69. Pokydko M, Oliinyk O, Tymchenko V: MEMS Gyroscope Based on MPU-6050 Sensor and ATmega328 Microcontroller. In: 2024 IEEE 7th International Conference on Smart Technologies in Power Engineering and Electronics (STEE): 24-26 Sept. 2024; 2024: TT3.39.31-TT33.39.36. Dang TH, Dang NH, Tran VT, Chung WY: A LoRaWAN-Based Smart Sensor Tag for Cow Behavior Monitoring. In: 2022 IEEE Sensors: 30 Oct.-2 Nov. 2022; 2022: 1-4. Catrett CC, Parsons IL, Dentinger JE, Norman DA, Webb SL, Stone AE, Street G, Karisch BB: PSII-12 Identifying behaviors and the ‘normal’ daily ethogram using accelerometers on grazing animals. Journal of Animal Science 2021, 99(Supplement_3):319-320. Li K, Fan D, Wu H, Zhao A: A new dataset for video-based cow behavior recognition. Scientific Reports 2024, 14(1):18702. Riaboff L, Aubin S, Bédère N, Couvreur S, Madouasse A, Goumand E, Chauvin A, Plantier G: Evaluation of pre-processing methods for the prediction of cattle behaviour from accelerometer data. Computers and Electronics in Agriculture 2019, 165:104961. Williams ML, James WP, Rose MT: Variable segmentation and ensemble classifiers for predicting dairy cow behaviour. Biosystems Engineering 2019, 178:156-167. Nogoy KM, Chon S-i, Park J-h, Sivamani S, Lee D-H, Choi SH: High Precision Classification of Resting and Eating Behaviors of Cattle by Using a Collar-Fitted Triaxial Accelerometer Sensor. In: Sensors. vol. 22; 2022. Pavlovic D, Czerkawski M, Davison C, Marko O, Michie C, Atkinson R, Crnojevic V, Andonovic I, Rajovic V, Kvascev G et al: Behavioural Classification of Cattle Using Neck-Mounted Accelerometer-Equipped Collars. In: Sensors. vol. 22; 2022. Arablouei R, Currie L, Kusy B, Ingham A, Greenwood PL, Bishop-Hurley G: In-situ classification of cattle behavior using accelerometry data. Computers and Electronics in Agriculture 2021, 183:106045. Uenishi S, Oishi K, Kojima T, Kitajima K, Yasunaka Y, Sakai K, Sonoda Y, Kumagai H, Hirooka H: A novel accelerometry approach combining information on classified behaviors and quantified physical activity for assessing health status of cattle: a preliminary study. Applied Animal Behaviour Science 2021, 235:105220. Yin Z, Liu L, Liu L, Zhang J, Wang Y: Dynamical recursive feature elimination technique for neurophysiological signal-based emotion recognition. Cognition, Technology & Work 2017, 19(4):667-685. Shekar BH, Dagnew G: Grid Search-Based Hyperparameter Tuning and Classification of Microarray Cancer Data. In: 2019 Second International Conference on Advanced Computational and Communication Paradigms (ICACCP): 25-28 Feb. 2019; 2019: 1-8. Jiménez A, Bautista F, Galina CS, Romero JJ, Rubio I: Behavioral Characteristics of Bos indicus Cattle after a Superovulatory Treatment Compared to Cows Synchronized for Estrus. Asian-Australas J Anim Sci 2011, 24(10):1365-1371. Lozada CC, Park RM, Daigle CL: Evaluating accurate and efficient sampling strategies designed to measure social behavior and brush use in drylot housed cattle. PLOS ONE 2023, 18(1):e0278233. Hirata M, Nakayama Y, Tobisa M: Interindividual variability in feeding station behavior in cattle: A preliminary study. Grassland Science 2010, 56(2):108-115. Tamura T, Okubo Y, Deguchi Y, Koshikawa S, Takahashi M, Chida Y, Okada K: Dairy cattle behavior classifications based on decision tree learning using 3-axis neck-mounted accelerometers. Animal Science Journal 2019, 90(4):589-596. Sprinkle JE, Sagers JK, Hall JB, Ellison MJ, Yelich JV, Brennan JR, Taylor JB, Lamb JB: Predicting Cattle Grazing Behavior on Rangeland using Accelerometers. Rangeland Ecology & Management 2021, 76:157-170. Kleanthous N, Hussain A, Mason A, Sneddon J: Data Science Approaches for the Analysis of Animal Behaviours. In: Intelligent Computing Methodologies: 2019// 2019; Cham: Springer International Publishing; 2019: 411-422. Kleanthous N, Hussain A, Khan W, Sneddon J, Mason A: Feature Extraction and Random Forest to Identify Sheep Behavior from Accelerometer Data. In: Intelligent Computing Methodologies: 2020// 2020; Cham: Springer International Publishing; 2020: 408-419. Breiman L: Random Forests. Machine Learning 2001, 45(1):5-32. Ibrahim T, Isaac KB, Francis B, Lule E, Hellen N, Chongomweru H, Marvin G: Interpretable Machine Learning Techniques for Predictive Cattle Behavior Monitoring. In: 2024 2nd International Conference on Sustainable Computing and Smart Systems (ICSCSS): 10-12 July 2024; 2024: 1219-1224. Daker M, Elsayaad F, Atia A: The Classification Of Cattle Behaviors Using Deep Learning. In: 2024 6th International Conference on Computing and Informatics (ICCI): 6-7 March 2024 2024; 2024: 28-33. Arcidiacono C, Porto SMC, Mancino M, Cascone G: Development of a threshold-based classifier for real-time recognition of cow feeding and standing behavioural activities from accelerometer data. Computers and Electronics in Agriculture 2017, 134:124-134. Hosseininoorbin S, Layeghy S, Kusy B, Jurdak R, Bishop-Hurley GJ, Greenwood PL, Portmann M: Deep learning-based cattle behaviour classification using joint time-frequency data representation. Computers and Electronics in Agriculture 2021, 187:106241. Gao G, Wang C, Wang J, Lv Y, Li Q, Ma Y, Zhang X, Li Z, Chen G: CNN-Bi-LSTM: A Complex Environment-Oriented Cattle Behavior Classification Network Based on the Fusion of CNN and Bi-LSTM. In: Sensors. vol. 23; 2023. Peng Y, Kondo N, Fujiura T, Suzuki T, Wulandari, Yoshioka H, Itoyama E: Classification of multiple cattle behavior patterns using a recurrent neural network with long short-term memory and inertial measurement units. Computers and Electronics in Agriculture 2019, 157:247-253. Table 2 Table 2 is available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files Table2.docx SupplementaryTable1.docx Cite Share Download PDF Status: Published Journal Publication published 20 Nov, 2025 Read the published version in BMC Veterinary Research → Version 1 posted Editorial decision: Revision requested 18 Aug, 2025 Reviews received at journal 15 Aug, 2025 Reviews received at journal 12 Aug, 2025 Reviewers agreed at journal 12 Aug, 2025 Reviewers agreed at journal 04 Aug, 2025 Reviewers agreed at journal 22 Jul, 2025 Reviewers invited by journal 28 May, 2025 Editor assigned by journal 21 May, 2025 Submission checks completed at journal 21 May, 2025 First submitted to journal 16 May, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6682405","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":462922773,"identity":"9a65a12e-ee6d-4298-a915-addcba4b4b6e","order_by":0,"name":"Khamta Pongsanun","email":"","orcid":"","institution":"Chulalongkorn University","correspondingAuthor":false,"prefix":"","firstName":"Khamta","middleName":"","lastName":"Pongsanun","suffix":""},{"id":462922776,"identity":"d49f7f6a-6217-4a22-9a6e-3bff68be6659","order_by":1,"name":"Tadsorn Apirak","email":"","orcid":"","institution":"Chulalongkorn University","correspondingAuthor":false,"prefix":"","firstName":"Tadsorn","middleName":"","lastName":"Apirak","suffix":""},{"id":462922778,"identity":"b895518f-9623-4ccf-827d-749da8099241","order_by":2,"name":"Leklerdsiriwong Aekaluck","email":"","orcid":"","institution":"Chulalongkorn University","correspondingAuthor":false,"prefix":"","firstName":"Leklerdsiriwong","middleName":"","lastName":"Aekaluck","suffix":""},{"id":462922783,"identity":"d9a704e7-a572-41af-be6c-0afe60d66c89","order_by":3,"name":"Sanphet Chunithipaisan","email":"","orcid":"","institution":"Chulalongkorn University","correspondingAuthor":false,"prefix":"","firstName":"Sanphet","middleName":"","lastName":"Chunithipaisan","suffix":""},{"id":462922784,"identity":"5efa47a7-aa9f-45ec-9dd6-e89673af0845","order_by":4,"name":"Chaidate Inchaisri","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA+UlEQVRIiWNgGAWjYDACCQaGAzwgBjPzAWYGBguIKGMDQS0GQC1sCcxgLggcJKCFAayFgceAOC38s5sfHnhT80eOgZ3n4+fCNgl5BvbDD5g/7sBjyZ1jBgfnHDMwZmDm3Sw9s03CsIEnzYDh4BncWgwkEgwO87AZJDYw825j5m2TAHo8B+iwNnxa0j8c5vlnUN/AzPMMpMW+gf8NIS05Bod52wwSGJh52EBaEhskCNgicSOn4ODcPmPDNmY2Y2mecxLJbRLPDA6cxaOFf0b65g9vvsnJ8/MffviZp8zGtp8/+eGDSjxa4IANmXGACA2jYBSMglEwCvAAACYQSVzung3jAAAAAElFTkSuQmCC","orcid":"","institution":"Chulalongkorn University","correspondingAuthor":true,"prefix":"","firstName":"Chaidate","middleName":"","lastName":"Inchaisri","suffix":""}],"badges":[],"createdAt":"2025-05-16 16:38:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6682405/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6682405/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12917-025-05092-1","type":"published","date":"2025-11-20T15:57:52+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":83629131,"identity":"a2ad5c53-f1d1-4dc9-9102-b98e8512af11","added_by":"auto","created_at":"2025-05-29 18:10:26","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1004737,"visible":true,"origin":"","legend":"\u003cp\u003ePrototype of the Tri-axial accelerometer and gyroscope sensor (Left) and placement on the cow (Right).\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6682405/v1/7163f713221988bb2ed9926f.png"},{"id":83629132,"identity":"6e457158-57a5-4f76-9b5e-10891fe41755","added_by":"auto","created_at":"2025-05-29 18:10:26","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":780141,"visible":true,"origin":"","legend":"\u003cp\u003eConceptual framework of this study\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6682405/v1/5c9cade50df92f29e772c0bd.jpg"},{"id":83629231,"identity":"0fd6a6a0-37aa-4cc8-95f4-2e5872fe34f6","added_by":"auto","created_at":"2025-05-29 18:18:26","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":3671172,"visible":true,"origin":"","legend":"\u003cp\u003eAcceleration signal patterns over a 24-hour period for seven individual cows.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6682405/v1/000cf6d600c87ff6d4537377.png"},{"id":83629133,"identity":"7a368378-07fe-40de-a668-bd865ac8126d","added_by":"auto","created_at":"2025-05-29 18:10:26","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":992619,"visible":true,"origin":"","legend":"\u003cp\u003eGyroscope signal patterns (GyroX, GyroY, GyroZ) recorded over a 24-hour period for seven individual cows.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6682405/v1/1e7e0d2e8d5d4aa4699c90f1.png"},{"id":83629136,"identity":"7913d808-b821-4240-a4e7-47ae5a68ac34","added_by":"auto","created_at":"2025-05-29 18:10:26","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":993975,"visible":true,"origin":"","legend":"\u003cp\u003eSignal Vector Magnitude (SVM_acc) from accelerometer data \u0026nbsp;overs a 24-hour period for seven individual cows.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6682405/v1/6f0fc4895175a9b31cce2ec4.png"},{"id":83629134,"identity":"0cc1a3ff-4fb6-4513-8175-4a603bbaa7a7","added_by":"auto","created_at":"2025-05-29 18:10:26","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":826894,"visible":true,"origin":"","legend":"\u003cp\u003eSignal Vector Magnitude (SVM_gyro) from gyroscope data \u0026nbsp;over a 24-hour period for seven individual cows.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-6682405/v1/a4da3bbfb0d1ec84a2d5c783.png"},{"id":96650290,"identity":"37369a9c-d9ff-4348-88bd-882a312f5d51","added_by":"auto","created_at":"2025-11-24 16:10:54","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":11849753,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6682405/v1/5d9c6252-0807-480a-9139-ad59fc2108e7.pdf"},{"id":83629230,"identity":"f9cc7e7d-0df4-449d-af5c-bf5a44b2c582","added_by":"auto","created_at":"2025-05-29 18:18:26","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":17047,"visible":true,"origin":"","legend":"","description":"","filename":"Table2.docx","url":"https://assets-eu.researchsquare.com/files/rs-6682405/v1/3cc544fb9b3c9d3bf65ccd17.docx"},{"id":83629128,"identity":"9c36cfb2-57a5-4c8d-bd4c-ac1837609bf3","added_by":"auto","created_at":"2025-05-29 18:10:26","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":26790,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTable1.docx","url":"https://assets-eu.researchsquare.com/files/rs-6682405/v1/2b422d3bbcf7d37383071350.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Classification of Individual Dairy Cow Behaviors Using Accelerometer, Gyroscope, and Integrated Sensor Models","fulltext":[{"header":"Background","content":"\u003cp\u003eMonitoring livestock behavior is a fundamental aspect of precision dairy farming, supporting animal welfare, early disease detection, and productivity improvements [1-3]. Accurate behavior monitoring facilitates the early detection of health and reproductive conditions, such as lameness, metabolic disorders, and estrus, thereby enhancing herd management outcomes \u0026nbsp;[2, 4, 5]. Traditional observation methods, such as visual scoring, are labor-intensive, subjective, and limited in both temporal resolution and scalability, which restricts their utility in modern herd management systems [6]. As an alternative, wearable sensor technologies, particularly accelerometers, have emerged as reliable tools for continuous and automated behavior monitoring in cattle [7-9]. Accelerometers are particularly effective at detecting postural changes and linear motions associated with behaviors such as lying, standing, or walking [9, 10]. However, accelerometers have limited ability to detect nuanced or rotational behaviors, such as transitions and head movements, due to their sensitivity being restricted to linear acceleration [9, 11]. In contrast, gyroscopes, which measure angular velocity, provide complementary information by capturing rotational movement data and have shown potential to enhance behavior classification, particularly for complex or dynamic behaviors such as walking [9, 11, 12]. These automated systems are particularly valuable in pasture-based or labor-constrained farming environments, where continuous manual observation is often infeasible [13, 14]. With the growing adoption of individualized precision management in dairy systems, behavior classification at the individual cow level using multimodal sensor data is becoming increasingly important for enabling targeted interventions and improving health and performance [2, 15].\u003c/p\u003e\n\u003cp\u003eNumerous studies have validated the effectiveness of accelerometers in detecting fundamental cattle behaviors such as lying, standing, and eating. For instance, tri-axial accelerometers could accurately classify lying and standing behaviors, though walking and transitions were more difficult to detect [9]. Similarly, high precision and sensitivity in classifying resting and eating behaviors using neck- and leg-mounted accelerometers have been reported [16]. However, accelerometers primarily capture linear motion and may struggle to detect rotational or complex movements, potentially leading to the misclassification of behaviors such as transitions between lying and standing [9, 16]. To address these limitations, recent studies have investigated the integration of gyroscope data, which measures angular velocity, to enhance behavior classification models. Machine learning algorithms applied to collar-mounted gyroscopes have demonstrated higher classification accuracy than accelerometers alone, with classification performance reaching up to 99% for specific cattle behaviors [17].\u003c/p\u003e\n\u003cp\u003eAlthough these findings are promising, the standalone and integrated use of gyroscope data remains insufficiently explored under real-world farming conditions, where factors such as sensor displacement, environmental interference, and individual variability may compromise data integrity [12]. Moreover, most existing classification models aggregate data across multiple animals, potentially obscuring individual variability in movement patterns. Such aggregation can mask subtle, cow-specific behavioral patterns, ultimately reducing the precision and applicability of classification models [18, 19]. Few studies have systematically evaluated and compared the performance of accelerometer-only, gyroscope-only, and combined sensor models at the individual-cow level, revealing a significant gap in the literature. Addressing this gap is essential for advancing individual-based monitoring systems capable of accurately interpreting the unique behavioral patterns of each animal, thereby improving animal welfare and farm management practices [16].\u003c/p\u003e\n\u003cp\u003eThis study aims to evaluate the performance of accelerometer, gyroscope, and combined sensor configurations for classifying cattle behaviors at the individual animal level. It addresses a critical knowledge gap by examining the distinct signal characteristics of each axis (X, Y, Z) from tri-axial accelerometers and gyroscopes in relation to four primary behaviors: lying, standing, eating, and walking. To assess the classification performance, models are developed and compared using three sensor input strategies: accelerometer-only, gyroscope-only, and a combination of both. A Random Forest classifier is employed due to its robustness in handling noisy, high-dimensional data and its demonstrated effectiveness in livestock behavior classification tasks [17, 20]. By conducting the analysis at the individual animal level rather than aggregating data across multiple animals, this research provides refined insights into movement-based behavioral classification. The findings are expected to inform the development of more precise and adaptable monitoring systems tailored to individual cattle, thereby advancing individualized livestock management.\u003c/p\u003e\n"},{"header":"Material and Methods","content":"\u003cp\u003e\u003cstrong\u003e1.\u0026nbsp; \u0026nbsp;Animal Management\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was conducted over a three-month period (August to October 2021) at the loose barn of the Farm Animal Hospital, Faculty of Veterinary Science, Chulalongkorn University, Nakhon Pathom, Thailand. Seven heifers (75\u0026ndash;87.5% Holstein-Friesian) were housed in a 30 \u0026times; 15 m concrete-floored enclosure designed to allow free movement and expression of natural behaviors. The cows were fed a total mixed ration comprising 80% roughage and 20% concentrate (on a dry matter basis) at 2.5% of body weight, provided twice daily at 09:00 and 14:00 h. Clean drinking water was available ad libitum throughout the study period.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.\u0026nbsp; \u0026nbsp;Sensor Configuration and Placement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAcceleration and angular velocity were continuously recorded over 90 days using a custom-built activity meter operating 24 hours per day. The device was developed in-house and comprised a tri-axis accelerometer and gyroscope sensor (MPU-6050, InvenSense Inc., California, USA), integrated with a wireless module, microcontroller, 3,700 mAh lithium battery, external antenna, and Wi-Fi router (Figure 1, left). The MPU-6050 includes an onboard digital motion processor capable of processing 6-axis motion data. The accelerometer and gyroscope featured full-scale measurement ranges of \u0026plusmn;2, \u0026plusmn;4, \u0026plusmn;8, and \u0026plusmn;16 g, and \u0026plusmn;250, \u0026plusmn;500, \u0026plusmn;1000, and \u0026plusmn;2000\u0026deg;/s, respectively [21].\u003c/p\u003e\n\u003cp\u003eThe sensor system was supported by a LoRa mainboard (Heltec Automation, Sichuan, China) for power management and wireless data transmission. Devices were enclosed in 3D-printed housings and securely mounted on the right side of each cow\u0026rsquo;s neck using adjustable collars (Figure 1, right) [22]. Antennas were positioned within 50 meters of the sensors to ensure consistent data communication with the gateway.\u003c/p\u003e\n\u003cp\u003eA dedicated computer, connected via LAN to a Wi-Fi router, served as the data collection hub. The system utilized XAMPP (Apache Friends, Berlin, Germany) for preliminary server testing and phpMyAdmin version 4.8.1 (phpMyAdmin Project, www.phpmyadmin.net) for SQL-based data management.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003cstrong\u003e3. \u0026nbsp; Data Processing and Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe overall workflow of this study comprises three main stages: data acquisition, data preparation, and classification model development. These stages represent a structured process that begins with the collection of synchronized sensor and behavioral data, continues with systematic data cleaning and feature preparation, and concludes with the development and evaluation of classification models using supervised machine learning techniques. As illustrated in Figure 2, this framework provides the basis for the detailed methodological descriptions presented in the following sections.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.1\u0026nbsp;Data Acquisition\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSensor data were continuously recorded over a 90-day period and stored in time-series format with timestamp annotations. Simultaneously, cow behaviors were recorded using a closed-circuit television (CCTV) system operating at 15 frames per second. Synchronization between video footage and sensor data was achieved through precise timestamp alignment.\u003c/p\u003e\n\u003cp\u003eTwo trained observers independently annotated behaviors by reviewing 24-hour video recordings throughout the 90-day study. Each time window was labeled as lying, standing, eating, or walking based on a standardized ethogram [23, 24]. To ensure consistency, both observers underwent joint training and annotated a pilot dataset to harmonize behavior interpretation. Inter-observer reliability was assessed using Cohen\u0026rsquo;s Kappa on randomly selected video segments. A 5% random subset of the dataset, sampled across individual cows and time periods, was annotated by both observers to ensure behavioral and temporal diversity. Inter-observer reliability, assessed using Cohen\u0026rsquo;s Kappa, was 0.84, indicating strong consistency. Discrepancies were resolved through discussion and consensus meetings. Annotation was conducted blind to the study\u0026rsquo;s hypotheses and model development, in line with best practices outlined in recent video-based annotation studies.\u003c/p\u003e\n\u003cp\u003eSegments containing artifacts, missing values, or overlapping behaviors were excluded from analysis. Although seven behaviors were initially annotated, only four including lying, standing, eating, and walking were retained for model development. Drinking was excluded due to an insufficient number of observed instances, while ruminating and other were removed because of substantial overlap with other behaviors and ambiguous neck movements that reduced classification reliability. Consequently, the final dataset included only four clearly distinguishable and mutually exclusive behaviors, as defined in Table 1.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 1. Definition of cow behaviors.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.625%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBehavior\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84.375%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDefinition\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.625%;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84.375%;\"\u003e\n \u003cp\u003eThe cow is in a resting posture with the ventral body surface in contact with the ground, supported by the sternum and one or both thighs. The neck is positioned either vertically or horizontally and may be flexed backward toward the hindquarters. Lateral recumbency, in which the cow lies fully on its side, was excluded from this category to maintain consistency in posture-based labeling.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.625%;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84.375%;\"\u003e\n \u003cp\u003eThe cow remains upright, supported by at least three legs, without forward or backward movement. The neck is aligned along the vertical axis, although minor movements related to comfort or social interactions may occur.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.625%;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84.375%;\"\u003e\n \u003cp\u003eThe cow stands in the feeding area on at least three legs with its head lowered into the feed bunk to ingest or masticate feed. The behavior is considered to end when the cow raises its head and maintains that position for at least five consecutive seconds.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.625%;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84.375%;\"\u003e\n \u003cp\u003eThe cow exhibits progressive movement, either forward or backward, covering more than two feet. The behavior involves sequential limb movements, with the head generally held in an upright position.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e3.2\u0026nbsp;Data Preprocessing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSensor data preprocessing was performed using Python within a Jupyter Notebook environment. The workflow included data import, inspection, cleaning, noise filtering, and feature extraction. Prior to automated processing, raw data were manually reviewed to ensure format consistency and structural completeness.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2.1\u0026nbsp; \u0026nbsp;\u0026nbsp;Data Cleaning\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEach data window was designed as a fixed 10-second interval. To ensure temporal consistency, segments with substantial timestamp deviations, indicative of signal loss or transmission errors, were excluded from analysis. A threshold of \u0026plusmn;1 second from the expected window duration was used to identify these irregular segments, following established practices [20, 25]. This filtering step minimized artifacts and improved the reliability of downstream feature extraction. The proportion of excluded segments was minimal and did not substantially affect the overall sample size or behavioral class distribution.\u003c/p\u003e\n\u003cp\u003eMissing values, malformed entries, and extreme outliers were removed using null value filtering functions, consistent with best practices in wearable sensor data analysis [20, 25]. To reduce signal noise, a rolling average filter was applied to smooth the time-series data prior to feature extraction.\u003c/p\u003e\n\u003cp\u003ePrior to feature extraction, the dataset was segmented according to predefined behavioral labels. This ensured that only stable, single-behavior intervals were included, minimizing the influence of behavioral transitions, and enhancing classification accuracy [25-27].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2.2\u0026nbsp; \u0026nbsp;\u0026nbsp;Feature Extraction and Selection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFollowing preprocessing, sensor data were segmented into 10-second windows, from which 79 descriptive features were extracted to characterize the movement patterns of dairy cows. These features included time-domain statistical metrics such as mean, standard deviation, minimum, maximum, sum, root mean square (RMS), kurtosis, interquartile range (IQR), and zero-crossing rate, as well as complexity and magnitude descriptors like entropy, signal magnitude area (SMA), and signal vector magnitude (SVM) [28, 29]. Entropy was calculated using scipy.stats.entropy, while SMA and SVM were derived from Euclidean combinations of tri-axial sensor data. Movement-specific indicators including dynamic body acceleration on individual axes (DBAX, DBAY, DBAZ), overall dynamic body acceleration (ODBA), and vectorial dynamic body acceleration (VeDBA) were computed following established formulas [30].\u003c/p\u003e\n\u003cp\u003eIn total, 42 features were derived from the accelerometer and 37 from the gyroscope, utilizing both raw axis values and composite statistical measures. A detailed breakdown of feature categories and sensor-specific counts is provided in Table 2. To ensure signal continuity and feature integrity, any window containing missing or undefined values in one or more sensor axes was excluded prior to feature extraction [29]. Feature extraction and all related computations were implemented in Python (version 3.10), using NumPy, pandas, and SciPy.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003cstrong\u003e3.3\u0026nbsp;Classification Model Development\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo classify cow behaviors, three Random Forest (RF) models were developed using features derived from accelerometer data, gyroscope data, and their combination. To preserve class distribution, the labeled dataset was initially split using stratified sampling into training/validation (80%) and testing (20%) subsets. The training/validation set was further divided (80:20) for hyperparameter tuning and model selection.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFeature selection was conducted using Recursive Feature Elimination (RFE) with F1-score\u0026ndash;based cross-validation, implemented through scikit-learn. RFE was applied independently for each cow-specific dataset to account for individual variability in movement patterns, consistent with individualized modeling practices in livestock behavior research [29]. The optimal number of features for each cow was determined by identifying the performance plateau in F1-score trends during Recursive Feature Elimination [31].\u003c/p\u003e\n\u003cp\u003eModel implementation was carried out in Python using the scikit-learn library, with support from pandas, NumPy, matplotlib, and joblib. The Random Forest (RF) algorithm, well suited for high-dimensional classification tasks, constructs \u0026nbsp;an ensemble of decision trees, with final predictions determined by majority voting. \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eHyperparameters including the number of trees (n_estimators), maximum tree depth (max_depth), and minimum sample thresholds for node splitting (min_samples_split, min_samples_leaf) were optimized using grid search with 5-fold cross-validation. This approach enabled performance evaluation across rotating validation sets while mitigating overfitting. To ensure robustness, each model was evaluated over 10 independent iterations, and averaged performance metrics were reported. A fixed random seed was applied to each iteration to ensure reproducibility throughout the modeling process.\u003c/p\u003e\n\u003cp\u003eModel performance was evaluated \u0026nbsp;using standard multiclass classification metrics: accuracy, precision, sensitivity (recall), and F1-score. Each metric was calculated on a per-class basis using a one-vs-rest strategy and summarized through macro-averaging ensuring equal weighting across all behavior classes [32]. The following formulas were used to compute these metrics:\u003c/p\u003e\n\u003cp\u003ePrecision = TP / (TP + FP)\u003c/p\u003e\n\u003cp\u003eSensitivity (Recall) = TP / (TP + FN)\u003c/p\u003e\n\u003cp\u003eF1-Score = 2 \u0026times; (Precision \u0026times; Recall) / (Precision + Recall)\u003c/p\u003e\n\u003cp\u003eAccuracy = Correct predictions / Total predictions\u003c/p\u003e\n\u003cp\u003eWhere TP = true positives, FP = false positives, and FN = false negatives. To address class imbalance across behavior categories, the class_weight=\u0026apos;balanced\u0026apos;\u0026nbsp;parameter was performed during optimum feature selection and model training. Feature scaling was applied using StandardScaler within a pipeline structure to ensure consistency and compatibility with potential scale-sensitive model extensions. All modeling procedures were executed independently for each individual cow (Cow 1 to Cow 7) using a standardized and reproducible pipeline.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.4\u0026nbsp;Statistical Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo preserve animal-level variability and avoid bias from data pooling, all analyses were conducted at the individual cow level (i.e., on a cow-by-cow basis). This approach enabled evaluation of both signal characteristics and machine learning (ML) performance within each animal, rather than aggregating results across the population. Accordingly, statistical comparisons and performance metrics were stratified by individual to capture between-animal heterogeneity in sensor signal patterns and behavioral expression. Statistical analyses were performed to evaluate differences in raw sensor signals and machine learning (ML) performance across four behavioral categories: lying, standing, eating, and walking. The Shapiro\u0026ndash;Wilk test was used to assess normality and guide the selection of appropriate statistical tests. For normally distributed variables, one-way ANOVA followed by Tukey\u0026rsquo;s Honest Significant Difference (HSD) test was applied. When the assumption of normality was violated, the Kruskal\u0026ndash;Wallis test with Dunn\u0026rsquo;s post hoc comparisons (Bonferroni-adjusted) was used [33, 34]. Homogeneity of variance was assumed for ANOVA and assessed through visual inspection of distributional characteristics. Statistical significance was defined as p \u0026lt; 0.05. Group-level differences were annotated using letter-based significance groupings generated by the multcompLetters and cldList functions.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eComparisons were conducted across sensor types, signal axes (AccX, AccY, AccZ, GyroX, GyroY, GyroZ), as well as derived features such as signal magnitude area (SMA) and signal vector magnitude (SVM). Descriptive statistics including mean, standard deviation, range, median, and interquartile range were computed using R (version 4.3.1, RStudio Build 2023.06.1+524) with the dplyr, data.table, and rcompanion packages. To ensure reliable inference, only groups with a minimum of three valid observations were included in the analysis.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMachine learning (ML) performance metrics, including accuracy, precision, sensitivity, and F1-score, were compared across sensor types for each behavior using the same test-selection strategy described above. This stratified approach accounted for differences in classification difficulty among behaviors [20]. Axis-specific comparisons within each cow and behavior were conducted using ANOVA followed by Tukey\u0026rsquo;s HSD test, with significant group differences indicated by letter-based annotations [28, 30].\u003c/p\u003e\n\u003cp\u003eAll descriptive summaries, statistical results, and significance groupings were consolidated and exported using the openxlsx package. Statistical analyses were implemented in R using the FSA, rcompanion, multcompView, and ggplot2 packages.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eSample Description\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTable 3. Number of cow data and distribution of behavioral instances across individual animals and activity classes.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"569\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCow No.\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 324px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCount of Labeled Behaviors\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTotal\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLying\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eStanding\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eEating\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eWalking\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e45,051\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e33,269\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e6,648\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e5,950\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 72px;\"\u003e\n \u003cp\u003e90,918\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e100,830\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e99,042\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e16,450\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e6,908\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 72px;\"\u003e\n \u003cp\u003e223,230\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e49,762\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e49,162\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e6,224\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e7,768\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 72px;\"\u003e\n \u003cp\u003e112,916\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e95,752\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e76,764\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e21,640\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e6,125\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 72px;\"\u003e\n \u003cp\u003e200,281\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e18,872\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e23,625\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e2,356\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e2,528\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 72px;\"\u003e\n \u003cp\u003e47,381\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e40,178\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e31,559\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e9,936\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e2,589\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 72px;\"\u003e\n \u003cp\u003e84,262\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e11,004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e11,810\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e1,428\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e1,300\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 72px;\"\u003e\n \u003cp\u003e25,542\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003eNumber of Instances\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e361,449\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e325,231\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e64,682\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e33,168\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 72px;\"\u003e\n \u003cp\u003e784,530\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003eInstances (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e46.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e41.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003e8.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e4.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 72px;\"\u003e\n \u003cp\u003e100.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eA total of 1,061,891 observations were collected from seven dairy cows over 90-day period using accelerometer and gyroscope sensors. After excluding technical failures, missing values, and estrus-related behaviors (53,095 observations), 784,530 observations (83.06%) remained for analysis, representing 1,794.1 hours of synchronized sensor and video-annotated data. Five cows contributed 90.7% of the dataset, while the remaining two accounted for 9.3%. The distribution of behavioral observations across the four target classes including lying, eating, standing, and walking is presented in Table 3. The dataset also included overlapping behaviors, such as rumination, which occasionally occurred concurrently with other activities.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBehavioral Signal Patterns\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTri-axial accelerometer data revealed distinct signal profiles corresponding to different behavioral states in dairy cows. Lying behavior was associated with low and stable values across all axes, particularly on the AccY axis, and corresponded to minimal movement. Moderate signal fluctuations were observed during standing and eating, with eating showing slightly greater variation. Walking behavior occurred intermittently and did not consistently result in elevated signal magnitudes across all axes. These patterns, derived from continuous 24-hour recordings, are visualized in Figure 3, which displays annotated time-series plots across seven individual cows.\u003c/p\u003e\n\u003cp\u003eA clear diurnal rhythm was evident in the temporal distribution of behaviors. Walking and eating occurred more frequently during daylight hours (09:00\u0026ndash;17:00), while lying behavior predominated during nighttime, reflecting rest-activity cycles typical of dairy cattle. Despite general consistency in these trends, subtle individual differences in signal amplitude and behavior duration were observed across cows.\u003c/p\u003e\n\u003cp\u003eTri-axial gyroscope data described rotational signal patterns across behaviors in dairy cows. Angular velocity profiles varied across behaviors, with lying consistently associated with low values across all axes. Standing showed moderate gyroscopic variation, while eating exhibited broader and more irregular angular fluctuations, particularly along the GyroX and GyroY axes, where several high-amplitude peaks were observed. Walking appeared intermittently and did not consistently produce distinguishable increases in angular velocity. Among all axes, GyroY demonstrated the largest dynamic range, which indicates higher sensitivity to rotational activity. Figure 4 also illustrates inter-animal variability in signal amplitude, suggesting cow-specific movement profiles within the same behavior.\u003c/p\u003e\n\u003cp\u003eThe signal vector magnitude derived from accelerometer data (SVM_acc) showed varying patterns across behaviors over a 24-hour period (Figure 5). Lying behavior consistently produced low SVM_acc values. Eating behavior was associated with irregular and elevated peaks in the signal. Standing behavior had moderate SVM_acc values with variable signal patterns, occasionally overlapping both the lower levels typical of lying and higher values seen in eating. Walking occurred intermittently and was not associated with clearly distinguishable changes in SVM_acc. Overall, SVM_acc patterns were similar across cows, although Cow 3 showed relatively higher variation in signal amplitude compared to others.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Gyroscope-based Signal Vector Magnitude (SVM_gyro) showed changes in signal profiles across behaviors and cows. Lying behavior was consistently associated with low SVM_gyro values across all individuals. Elevated SVM_gyro values were observed during eating periods, with some instances exceeding 60\u0026deg;/s. Walking occurred sporadically and did not consistently produce elevated SVM_gyro levels. SVM_gyro profiles were generally similar across cows, although Cows 3 and 4 showed relatively larger variations in signal values, while Cows 2 and 5 displayed lower SVM_gyro amplitudes. These patterns are illustrated in Figure 6, which presents 24-hour SVM_gyro time-series plots for each individual cow.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003cstrong\u003eSensor Signal Distribution\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTable 4. Descriptive statistics of Signal Vector Magnitude from accelerometer (SVM_acc) and gyroscope (SVM_gyro) across individual cows behaviors.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"666\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCow\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBehavior\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 210px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSVM_acc\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 228px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSVM_gyro\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 108px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean\u0026plusmn;SE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin-Max\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 114px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean\u0026plusmn;SE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 114px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin-Max\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003eCow No.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e45,051\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.33\u0026plusmn;0.19\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.01-2.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e7.34\u0026plusmn;8.84\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.01-79.48\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e33,269\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.31\u0026plusmn;0.2\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0-2.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e18.82\u0026plusmn;14.87\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.07-80.12\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e6,648\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.45\u0026plusmn;0.19\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.02-1.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e28.93\u0026plusmn;13.79\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e1.12-80.12\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e5,950\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.31\u0026plusmn;0.2\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.01-1.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e29.36\u0026plusmn;14.12\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.46-80.12\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003eCow No.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e100,830\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.37\u0026plusmn;0.2\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0-1.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e8.84\u0026plusmn;12.23\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.01-153.53\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e99,042\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.31\u0026plusmn;0.21\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0-2.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e21.13\u0026plusmn;17.59\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.02-353.92\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e16,450\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.55\u0026plusmn;0.22\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.02-2.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e33.93\u0026plusmn;21.15\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.04-371.51\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e6,908\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.3\u0026plusmn;0.19\u003csup\u003ed\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.01-1.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e36.1\u0026plusmn;21.63\u003csup\u003ed\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.19-313.33\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003eCow No.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e49,762\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.33\u0026plusmn;0.15\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.01-1.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e6.56\u0026plusmn;8.94\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.02-78.39\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e49,162\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.38\u0026plusmn;0.18\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.01-1.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e20.41\u0026plusmn;16.37\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.03-81.61\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e6,224\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.54\u0026plusmn;0.19\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.03-2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e32.58\u0026plusmn;15.12\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e1.57-81.23\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e7,768\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.38\u0026plusmn;0.18\u003csup\u003ed\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.03-1.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e35.51\u0026plusmn;14.56\u003csup\u003ed\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.6-81.51\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003eCow No.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e95,752\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.29\u0026plusmn;0.16\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0-1.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e4.99\u0026plusmn;6.41\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.01-135\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e76,764\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.31\u0026plusmn;0.18\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.01-2.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e22.5\u0026plusmn;20.11\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.08-135\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e21,640\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.48\u0026plusmn;0.18\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.01-2.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e28.51\u0026plusmn;18.7\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.76-135\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e6,125\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.36\u0026plusmn;0.2\u003csup\u003ed\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.03-2.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e35.22\u0026plusmn;25.98\u003csup\u003ed\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.63-135\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003eCow No.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e18,872\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.26\u0026plusmn;0.14\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.01-1.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e8.34\u0026plusmn;14.26\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.04-181.6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e23,625\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.28\u0026plusmn;0.17\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0-1.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e32.91\u0026plusmn;29.92\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.14-193.7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e2,356\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.47\u0026plusmn;0.18\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.02-1.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e52.9\u0026plusmn;38.97\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.56-192.98\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e2,528\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.33\u0026plusmn;0.17\u003csup\u003ed\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.02-1.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e48.28\u0026plusmn;34.04\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e1.81-192.33\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003eCow No.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e40,178\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.28\u0026plusmn;0.16\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0-1.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e5.27\u0026plusmn;6.2\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.05-52.69\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e31,559\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.31\u0026plusmn;0.18\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.01-1.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e21.71\u0026plusmn;15.13\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.22-52.69\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e9,936\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.48\u0026plusmn;0.17\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.04-2.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e27.25\u0026plusmn;13.52\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.76-52.69\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e2,589\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.35\u0026plusmn;0.2\u003csup\u003ed\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.03-2.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e30.9\u0026plusmn;14.64\u003csup\u003ed\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.63-52.69\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003eCow No.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e11,004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.3\u0026plusmn;0.15\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.03-1.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e4.09\u0026plusmn;6.08\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.02-40.76\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e11,810\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.25\u0026plusmn;0.18\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0-1.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e15.96\u0026plusmn;11.98\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e0.04-40.88\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e1,428\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.5\u0026plusmn;0.18\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.11-1.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e28.37\u0026plusmn;9.19\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e3.63-40.88\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e1,300\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 108px;\"\u003e\n \u003cp\u003e0.34\u0026plusmn;0.18\u003csup\u003ed\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.03-1.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e28.77\u0026plusmn;9.5\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 114px;\"\u003e\n \u003cp\u003e3.08-40.88\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eDifferent lowercase superscript letters (a\u0026ndash;d) indicate statistically significant differences (p \u0026lt; 0.05) of mean values among behaviors.\u003c/p\u003e\n\u003cp\u003eDescriptive analysis of Signal Vector Magnitude from accelerometer (SVM_acc) and gyroscope (SVM_gyro) data revealed clear behavioral differentiation across the four target activities. Lying behavior consistently produced the lowest mean values for both SVM_gyro and SVM_acc, reflecting minimal body movement and rotational activity, although for SVM_acc, this pattern was observed in five out of seven cows. \u0026nbsp; In contrast, eating produced the highest average SVM values, driven by dynamic head and neck motion. Walking exhibited similar SVM_gyro levels to eating in several cows, indicating comparable rotational intensity during these behaviors. Standing, as expected, presented intermediate values.\u003c/p\u003e\n\u003cp\u003eStatistical comparisons, conducted at the individual-animal level and reported in Table 4, confirmed significant differences between behaviors (p \u0026lt; 0.05), as denoted by differing lowercase subscripts. However, in Cows 1, 5, and 7, no significant difference was detected between eating and walking for SVM_gyro, suggesting overlapping motion patterns in these more active states. These findings emphasize both the discriminative power of SVM features and the presence of cow-specific variation in behavior-related movement dynamics.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 5. Descriptive statistics of raw tri-axial accelerometer signals (AccX, AccY, AccZ) across behaviors in individual cows.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"702\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCow\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBehavior\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccX\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 169px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccY\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 167px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccZ\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean\u0026plusmn;SE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin-Max\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean\u0026plusmn;SE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin-Max\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean\u0026plusmn;SE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin-Max\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003eCow No.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e45,051\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.13\u0026plusmn;0.28\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.4-2.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.08\u0026plusmn;0.09\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.49-0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.07\u0026plusmn;0.19\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-0.96-1.04\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e33,269\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.04\u0026plusmn;0.29\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.94-1.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.07\u0026plusmn;0.09\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.49-0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0\u0026plusmn;0.18\u003csup\u003eb,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.05-1.14\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e6,648\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.3\u0026plusmn;0.26\u003csup\u003ed,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.42-1.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.12\u0026plusmn;0.13\u003csup\u003ed,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.47-0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.11\u0026plusmn;0.21\u003csup\u003ed,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.06-1.14\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e5,950\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.14\u0026plusmn;0.24\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.67-1.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.06\u0026plusmn;0.11\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.49-0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.06\u0026plusmn;0.2\u003csup\u003ec,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.06-1.14\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003eCow No.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e100,830\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.14\u0026plusmn;0.21\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-0.83-1.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.07\u0026plusmn;0.09\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.96-0.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.16\u0026plusmn;0.28\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.16-1.2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e99,042\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.02\u0026plusmn;0.21\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-2.1-2.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.06\u0026plusmn;0.09\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-1.31-1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.16\u0026plusmn;0.24\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-2-1.57\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e16,450\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.31\u0026plusmn;0.25\u003csup\u003ed,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-2-1.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.15\u0026plusmn;0.12\u003csup\u003ed,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-1.31-0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.31\u0026plusmn;0.25\u003csup\u003ec,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-2-0.78\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e6,908\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.08\u0026plusmn;0.2\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.61-1.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.05\u0026plusmn;0.11\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-1.31-1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.1\u0026plusmn;0.23\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.52-1.54\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003eCow No.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e49,762\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.09\u0026plusmn;0.19\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-0.7-1.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.05\u0026plusmn;0.06\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.38-0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.19\u0026plusmn;0.2\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.39-0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e49,162\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.01\u0026plusmn;0.28\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.39-1.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.07\u0026plusmn;0.1\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.38-0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.24\u0026plusmn;0.17\u003csup\u003eb,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.39-0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e6,224\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.29\u0026plusmn;0.26\u003csup\u003ed,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.39-1.43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.13\u0026plusmn;0.13\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.38-0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.34\u0026plusmn;0.18\u003csup\u003ed,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.39-0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e7,768\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.04\u0026plusmn;0.24\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.39-1.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.06\u0026plusmn;0.12\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.38-0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.27\u0026plusmn;0.17\u003csup\u003ec,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.38-0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003eCow No.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e95,752\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.12\u0026plusmn;0.22\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.67-1.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.06\u0026plusmn;0.07\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.38-0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.07\u0026plusmn;0.18\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.22-0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e76,764\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.03\u0026plusmn;0.26\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.91-1.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.06\u0026plusmn;0.08\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.38-0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.17\u0026plusmn;0.15\u003csup\u003eb,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.22-0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e21,640\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.25\u0026plusmn;0.26\u003csup\u003ed,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.97-1.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.11\u0026plusmn;0.1\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.38-0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.28\u0026plusmn;0.17\u003csup\u003ed,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.22-0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e6,125\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.07\u0026plusmn;0.28\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.63-1.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.06\u0026plusmn;0.1\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.38-0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.21\u0026plusmn;0.17\u003csup\u003ec,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.22-0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003eCow No.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e18,872\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.11\u0026plusmn;0.2\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.14-0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.05\u0026plusmn;0.05\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.32-0.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.08\u0026plusmn;0.14\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-0.71-0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e23,625\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.08\u0026plusmn;0.22\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.37-0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.06\u0026plusmn;0.07\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.32-0.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.16\u0026plusmn;0.14\u003csup\u003eb,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-0.92-0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e2,356\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.28\u0026plusmn;0.24\u003csup\u003ed,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.13-0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.11\u0026plusmn;0.1\u003csup\u003ed,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.32-0.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.26\u0026plusmn;0.16\u003csup\u003ed,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-0.92-0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e2,528\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.16\u0026plusmn;0.21\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-0.9-0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.06\u0026plusmn;0.09\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.32-0.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.18\u0026plusmn;0.16\u003csup\u003ec,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-0.92-0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003eCow No.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e40,178\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.12\u0026plusmn;0.22\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.67-1.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.06\u0026plusmn;0.07\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.32-0.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.06\u0026plusmn;0.17\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.22-0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e31,559\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.04\u0026plusmn;0.25\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.67-1.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.06\u0026plusmn;0.08\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.32-0.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.18\u0026plusmn;0.15\u003csup\u003eb,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.22-0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e9,936\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.27\u0026plusmn;0.23\u003csup\u003ed,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.67-1.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.11\u0026plusmn;0.1\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.32-0.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.3\u0026plusmn;0.16\u003csup\u003ed,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.22-0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e2,589\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.07\u0026plusmn;0.26\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-1.63-1.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.06\u0026plusmn;0.1\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.32-0.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.22\u0026plusmn;0.16\u003csup\u003ec,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.22-0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003eCow No.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e11,004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.04\u0026plusmn;0.13\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-0.58-1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.05\u0026plusmn;0.06\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.31-0.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.23\u0026plusmn;0.18\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.39-0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e11,810\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.05\u0026plusmn;0.23\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-0.98-1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.05\u0026plusmn;0.08\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.31-0.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.09\u0026plusmn;0.16\u003csup\u003eb,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-1.48-1.16\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e1,428\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.34\u0026plusmn;0.19\u003csup\u003ed,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-0.98-0.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.14\u0026plusmn;0.11\u003csup\u003ed,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.31-0.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.24\u0026plusmn;0.21\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-0.98-0.59\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e1,300\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e-0.22\u0026plusmn;0.22\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e-0.98-0.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.07\u0026plusmn;0.12\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e-0.31-0.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-0.14\u0026plusmn;0.14\u003csup\u003ec,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e-0.61-0.49\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eDifferent lowercase superscript letters (a\u0026ndash;d) indicate statistically significant differences of raw tri-axial accelerometer signal among behaviors (p \u0026lt; 0.05). Uppercase superscript letters (A-C) indicate significant differences of raw tri-axial accelerometer signal among axes (X, Y, and Z) within each behavior (p \u0026lt; 0.05).\u003c/p\u003e\n\u003cp\u003eAnalysis of raw tri-axial accelerometer signals (AccX, AccY, AccZ) revealed distinct patterns across behavioral states, as detailed in Table 5. Lying and standing exhibited the lowest signal variation across all three axes, consistent with their classification as low-movement or static behaviors. In contrast, eating behavior was characterized by negative mean values and high standard deviations, particularly in AccX and AccY, indicating frequent and variable head and neck movements. Walking also showed elevated variability, though generally less pronounced than eating.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 6. Descriptive statistics of raw tri-axial gyroscope signals (GyroX, GyroY, GyroZ) across behaviors in individual cows.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"744\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCow\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBehavior\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGyrX\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGyrY\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGyrZ\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean\u0026plusmn;SE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin-Max\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean\u0026plusmn;SE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin-Max\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean\u0026plusmn;SE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin-Max\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eCow No.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e45,051\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.49\u0026plusmn;4.32\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-37.34-33.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.68\u0026plusmn;7.77\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-63.84-62.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-1.27\u0026plusmn;7.13\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-31.41-33.6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e33,269\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.27\u0026plusmn;10.72\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-37.34-33.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.89\u0026plusmn;18.65\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-63.84-62.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.67\u0026plusmn;10.55\u003csup\u003eb,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-31.41-33.6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e6,648\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.33\u0026plusmn;17.16\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-37.34-33.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e1.75\u0026plusmn;22.92\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-63.84-62.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.86\u0026plusmn;14.27\u003csup\u003eb,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-31.41-33.6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e5,950\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.22\u0026plusmn;15.62\u003csup\u003eab,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-37.34-33.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.99\u0026plusmn;24.75\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-63.84-62.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.75\u0026plusmn;14.27\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-31.41-33.6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eCow No.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e100,830\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.79\u0026plusmn;4.5\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-36.54-52.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.08\u0026plusmn;13.86\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-147.92-151.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.21\u0026plusmn;3.85\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-67.61-67.03\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e99,042\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e1.07\u0026plusmn;10.12\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-36.54-52.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-1.16\u0026plusmn;23.61\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-207.18-350.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.11\u0026plusmn;9.68\u003csup\u003eb,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-67.61-67.03\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e16,450\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e1.98\u0026plusmn;16.56\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-36.54-52.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-3.45\u0026plusmn;33.39\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-361.56-350.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.03\u0026plusmn;13.92\u003csup\u003eb,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-67.61-67.03\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e6,908\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e1.83\u0026plusmn;16.45\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-36.54-52.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-2.89\u0026plusmn;35.68\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-259.52-307.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.42\u0026plusmn;14.69\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-67.61-67.03\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eCow No.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e49,762\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.19\u0026plusmn;4.97\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-39.8-49.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.26\u0026plusmn;8.54\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-47.4-48.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.07\u0026plusmn;5.02\u003csup\u003ea,AB\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-44.36-49.2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e49,162\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.16\u0026plusmn;11.04\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-39.8-49.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.27\u0026plusmn;20.51\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-47.4-48.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.69\u0026plusmn;11.88\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-44.36-49.2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e6,224\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.29\u0026plusmn;16.3\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-39.8-49.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e3.89\u0026plusmn;27.14\u003csup\u003ed,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-47.4-48.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.71\u0026plusmn;16.5\u003csup\u003ed,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-44.36-49.2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e7,768\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.36\u0026plusmn;16.2\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-39.8-49.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-1.97\u0026plusmn;29.96\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-47.4-48.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e2.25\u0026plusmn;17.44\u003csup\u003ec,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-44.36-49.2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eCow No.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e95,752\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e1.17\u0026plusmn;3.69\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-58.24-57.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e2.68\u0026plusmn;6.03\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-112.12-114.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.15\u0026plusmn;2.73\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-42.48-41.78\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e76,764\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e1.2\u0026plusmn;13.38\u003csup\u003eab,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-58.24-57.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e1.99\u0026plusmn;24.79\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-112.12-114.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.01\u0026plusmn;10.57\u003csup\u003eb,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-42.48-41.78\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e21,640\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e1.41\u0026plusmn;17.55\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-58.24-57.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e2.23\u0026plusmn;26.78\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-112.12-114.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.1\u0026plusmn;11.41\u003csup\u003eab,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-42.48-41.78\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e6,125\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.37\u0026plusmn;19.77\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-58.24-57.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.37\u0026plusmn;36.21\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-112.12-114.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.66\u0026plusmn;14.59\u003csup\u003ec,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-42.48-41.78\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eCow No.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e18,872\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.39\u0026plusmn;6.6\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-73.1-71.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e1.31\u0026plusmn;13.7\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-152.39-166.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0\u0026plusmn;6.3\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-62.29-66.41\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e23,625\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.69\u0026plusmn;16.29\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-73.1-71.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e4.51\u0026plusmn;37.75\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-152.39-166.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-1.05\u0026plusmn;16.32\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-62.29-66.41\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e2,356\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e1.82\u0026plusmn;26.65\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-73.1-71.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e-0.91\u0026plusmn;55.06\u003csup\u003ed,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-152.39-166.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e4.61\u0026plusmn;23.48\u003csup\u003ec,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-62.29-66.41\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e2,528\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-2.4\u0026plusmn;23.31\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-73.1-71.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e7.68\u0026plusmn;48.46\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-152.39-166.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-1.68\u0026plusmn;23.04\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-62.29-66.41\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eCow No.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e40,178\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e1.35\u0026plusmn;3.67\u003csup\u003eab,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-24.08-26.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e2.98\u0026plusmn;5.95\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-36.94-42.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.15\u0026plusmn;2.58\u003csup\u003ea,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-16.65-16.46\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e31,559\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e1.29\u0026plusmn;12.4\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-24.08-26.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e2.53\u0026plusmn;21.33\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-36.94-42.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.09\u0026plusmn;9.13\u003csup\u003eab,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-16.65-16.46\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e9,936\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e1.61\u0026plusmn;16.19\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-24.08-26.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e2.55\u0026plusmn;23.47\u003csup\u003eab,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-36.94-42.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.07\u0026plusmn;10.17\u003csup\u003eb,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-16.65-16.46\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e2,589\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.58\u0026plusmn;16.53\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-24.08-26.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.85\u0026plusmn;27.61\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-36.94-42.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.52\u0026plusmn;11.5\u003csup\u003ec,C\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-16.65-16.46\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eCow No.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eLying\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e11,004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.04\u0026plusmn;3.58\u003csup\u003ea,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-20.09-19.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.04\u0026plusmn;5.81\u003csup\u003ea,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-30.79-31.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e0.28\u0026plusmn;2.64\u003csup\u003eab,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-14.78-15.8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eStanding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e11,810\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.62\u0026plusmn;8.91\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-20.09-19.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e1.87\u0026plusmn;15.77\u003csup\u003eb,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-30.79-31.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e1.31\u0026plusmn;8.03\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-14.78-15.8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eEating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e1,428\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.81\u0026plusmn;14.38\u003csup\u003eabc,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-20.09-19.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e4.99\u0026plusmn;23.26\u003csup\u003ec,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-30.79-31.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-0.19\u0026plusmn;10.86\u003csup\u003eb,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-14.78-15.8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eWalking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e1,300\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e-1.74\u0026plusmn;15.43\u003csup\u003ec,A\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-20.09-19.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 96px;\"\u003e\n \u003cp\u003e2.84\u0026plusmn;23.15\u003csup\u003ebc,B\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 90px;\"\u003e\n \u003cp\u003e-30.79-31.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 102px;\"\u003e\n \u003cp\u003e1.09\u0026plusmn;11.59\u003csup\u003eac,AB\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 78px;\"\u003e\n \u003cp\u003e-14.78-15.8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eDifferent lowercase superscript letters (a\u0026ndash;d) indicate statistically significant differences of raw tri-axial gyroscope signal among behaviors (p \u0026lt; 0.05). Uppercase superscript letters (A-C) indicate significant differences of raw tri-axial gyroscope signal among axes (X, Y, and Z) within each behavior (p \u0026lt; 0.05).\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; Tri-axial gyroscope signals (GyroX, GyroY, GyroZ) demonstrated clear behavioral distinctions in rotational movement patterns, as detailed in Table 6. Lying and standing were consistently associated with the lowest mean values and minimal variability across all axes, reflecting limited angular motion during these postural states. In contrast, eating and walking produced markedly higher gyroscopic activity, particularly along the GyroY and GyroZ axes, indicative of frequent head and body rotations.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel Performance\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTable 7. Classification performance of Random Forest models using accelerometer, gyroscope, and combined features across four behaviors.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"816\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBehavior\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDevice\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 165px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSensitivity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 171px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eF1-score\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOverall Accuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean \u0026plusmn; SE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin-Max\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean \u0026plusmn; SE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin-Max\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean \u0026plusmn; SE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin-Max\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean \u0026plusmn; SE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin-Max\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLying\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eAcc\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e93.33\u0026plusmn;2.45\u003csup\u003eab\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e90.75-96.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e94.52\u0026plusmn;1.6\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e92.01-96.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e93.91\u0026plusmn;1.93\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e91.55-96.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e89.44\u0026plusmn;3.5\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e86.26-94.15\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eGyr\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e92.93\u0026plusmn;2.07\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e90.39-96.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e94.5\u0026plusmn;1.91\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e91.9-96.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e93.71\u0026plusmn;1.88\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e91.14-96.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e88.35\u0026plusmn;3.63\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e83.46-93.56\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eCom\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e95.94\u0026plusmn;1.66\u003csup\u003ea\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e94.02-98.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e96.81\u0026plusmn;1.25\u003csup\u003eb\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e94.8-98.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e96.37\u0026plusmn;1.34\u003csup\u003eb\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e94.81-98.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e92.74\u0026plusmn;2.68\u003csup\u003eb\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e89.99-96.77\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eStanding\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eAcc\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e85.44\u0026plusmn;4.99\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e80.04-92.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e90.28\u0026plusmn;4.33\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e84.51-95.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e87.78\u0026plusmn;4.57\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e82.68-93.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e89.44\u0026plusmn;3.5\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e86.26-94.15\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eGyr\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e83.63\u0026plusmn;5.66\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e75.52-92.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e90.81\u0026plusmn;2.81\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e87.76-95.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e87.03\u0026plusmn;4.11\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e81.33-92.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e88.35\u0026plusmn;3.63\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e83.46-93.56\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eCom\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e89.63\u0026plusmn;3.9\u003csup\u003ea\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e84.4-95.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e93.67\u0026plusmn;3.02\u003csup\u003ea\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e89.72-97.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e91.59\u0026plusmn;3.34\u003csup\u003ea\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e88.21-96.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e92.74\u0026plusmn;2.68\u003csup\u003eb\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e89.99-96.77\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eEating\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eAcc\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e83.61\u0026plusmn;5.93\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e77.67-94.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e65.65\u0026plusmn;10.75\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e54.05-84.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e73.29\u0026plusmn;8.3\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e63.75-89.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e89.44\u0026plusmn;3.5\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e86.26-94.15\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eGyr\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e82.65\u0026plusmn;7.83\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e74.57-96.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e42.44\u0026plusmn;13.78\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e30.23-71.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e55.4\u0026plusmn;12.59\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e44.83-82.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e88.35\u0026plusmn;3.63\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e83.46-93.56\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eCom\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e86.29\u0026plusmn;4.53\u003csup\u003ea\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e83.64-95.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e72.46\u0026plusmn;8.44\u003csup\u003ea\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e62.16-87.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e78.65\u0026plusmn;6.57\u003csup\u003ea\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e71.32-91.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e92.74\u0026plusmn;2.68\u003csup\u003eb\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e89.99-96.77\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eWalking\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eAcc\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e90.65\u0026plusmn;7.35\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e82.06-100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e31.78\u0026plusmn;17.23\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e5.34-59.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e44.65\u0026plusmn;20.3\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e10.15-74.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e89.44\u0026plusmn;3.5\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e86.26-94.15\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eGyr\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e91.33\u0026plusmn;9.77\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e75.56-99.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e34.34\u0026plusmn;16.8\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e15-62.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e47.62\u0026plusmn;17.33\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e26.01-77.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e88.35\u0026plusmn;3.63\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e83.46-93.56\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eCom\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e93.26\u0026plusmn;5.96\u003csup\u003ea\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e83.64-100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e46.76\u0026plusmn;16.22\u003csup\u003ea\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e20.34-65.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e60.61\u0026plusmn;15.8\u003csup\u003ea\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e33.67-78.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e92.74\u0026plusmn;2.68\u003csup\u003eb\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e89.99-96.77\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eDifferent lowercase superscript letters (a\u0026ndash;d) indicate significant differences (p \u0026lt; 0.05) of mean classification performance among device types (accelerometer, gyroscope, combination), based on 70 data points (10 iterations per cow \u0026times; 7 cows). Bolded values represent the highest means for each performance metric across device types (accelerometer, gyroscope, and combination) within each behavioral classification.\u003c/p\u003e\n\u003cp\u003eClassification performance metrics including precision, sensitivity, F1-score, and overall accuracy varied across behaviors and sensor configurations, as summarized in Table 7. For lying, the combined-sensor model significantly outperformed both single-sensor models across all metrics (p \u0026lt; 0.05), demonstrating its advantage in detecting low-movement behaviors. In the case of standing, overall accuracy was significantly higher for the combined model, although differences in precision, sensitivity, and F1-score were not statistically significant. For eating, the combined model yielded significantly higher sensitivity and F1-score compared to the gyroscope-only model (p \u0026lt; 0.05), while precision did not differ significantly across configurations. Walking proved to be the most challenging behavior to classify, with all models showing relatively low sensitivity and F1-score. However, the combined model achieved significantly higher overall accuracy for this class, suggesting modest gains through sensor fusion even in more dynamic and variable behaviors.\u003c/p\u003e\n\u003cp\u003eIndividual-level results, presented in Supplementary Table 1, revealed notable inter-cow variability. Nevertheless, the combined model generally produced higher or more consistent performance across cows, particularly for lying and standing. Classification performance for eating and walking was more variable, both across sensor types and among individual cows, reflecting the complexity and heterogeneity of these behaviors.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study aimed to examine the characteristics of accelerometer and gyroscope signals across four behavioral categories (lying, standing, eating, and walking) and to evaluate the effect of different sensor configurations on classification performance in individual dairy cows. The analysis revealed clear behavioral distinctions in the sensor signal patterns. For instance, lying behavior was associated with lower signal variability, particularly in the gyroscope Z-axis and accelerometer Z-axis. In contrast, walking consistently exhibited higher signal intensity, especially in the accelerometer X-axis and gyroscope X- and Y-axes. Eating, however, was characterized by elevated and irregular signal amplitudes across both sensor types and multiple axes, notably in GyroY and AccX, reflecting complex head and neck movements.\u0026nbsp;\u0026nbsp;These findings underscore the potential of axis-specific features in distinguishing movement states. The 10-second window length used for segmentation was selected to balance temporal granularity with behavioral stability, allowing for representative movement capture while minimizing noise.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eRegarding classification performance, models utilizing both accelerometer and gyroscope data achieved the highest accuracy and F1-scores across all behaviors, outperforming those using accelerometer-only or gyroscope-only inputs. This advantage was particularly pronounced in behaviors such as lying and standing, which exhibited more subtle signal dynamics. Nevertheless, performance varied across individual cows and behaviors, particularly in eating and walking, reflecting the influence of cow-specific movement patterns and the inherent complexity of differentiating between dynamic behaviors. These results collectively highlight the importance of incorporating both translational and rotational motion data to enhance behavioral recognition in dairy cows.\u003c/p\u003e\n\u003cp\u003e1.\u0026nbsp; \u0026nbsp;Behavior-Specific Sensor Characteristics Across Different Axes\u003c/p\u003e\n\u003cp\u003eThe tri-axial analysis of accelerometer and gyroscope signals revealed distinct axis-specific sensitivities associated with different behavioral states.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFor accelerometer data, the Z-axis consistently exhibited low and stable values during lying, reflecting minimal vertical displacement, while eating behavior produced the largest deviations, especially along the X- and Y-axes, indicating active head and neck movement. Standing behavior exhibited intermediate signal values across all axes, while walking did not consistently produce elevated or distinctive signal magnitudes, suggesting it may be less distinguishable using raw accelerometer data alone. These findings align with previous studies indicating that posture-related behaviors typically generate clearer accelerometric patterns than locomotor activities [9, 10].\u003c/p\u003e\n\u003cp\u003eFor gyroscope signals, all three axes remained relatively stable during lying, with GyroZ showing the lowest variation across cows. In contrast, eating was characterized by the highest rotational intensity, particularly in the GyroY and GyroZ axes, suggesting active and frequent head rotations during this behavior. Standing behavior exhibited moderate angular variation, while walking produced variable, cow-dependent signal patterns that occasionally overlapped with eating in both amplitude and distribution. This is consistent with previous gyroscope-based studies, which have shown that rotational movement patterns are more effective for detecting feeding and transition behaviors than for identifying consistent walking signatures [30]. Notably, signal variability differed among cows even within the same behavior, highlighting inter-individual differences in movement patterns. These may reflect differences in body conformation, locomotion patterns, or environmental conditions such as pen structure or feeding space. Such inter-animal variation has been noted in earlier sensor-based studies and underscores the value of personalized models in behavior monitoring [9, 35]. These cow-specific patterns were evident in both signal amplitude and axis distribution, reinforcing the importance of individualized analysis in behavior classification systems.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Overall, these findings underscore the importance of directional sensitivity in interpreting behavioral signals. Axis-specific features, particularly from GyroY and AccX/AccY during active behaviors, offer valuable discriminative power and should be prioritized in sensor placement strategies and feature selection processes for future precision livestock monitoring applications [28, 36, 37].\u003c/p\u003e\n\u003cp\u003e2.\u0026nbsp; \u0026nbsp;Classification Performance and Model Behavior\u003c/p\u003e\n\u003cp\u003eThe integration of accelerometer and gyroscope data consistently produced superior classification performance across behaviors and individual cows, highlighting the advantages of sensor fusion for behavioral modeling. This enhancement is likely attributable to the complementary nature of translational and rotational data streams, wherein accelerometers effectively capture posture-related changes, and gyroscopes provide critical information on dynamic orientation, particularly during locomotor activities. Similar improvements in classification performance achieved through sensor fusion have been reported in recent precision livestock studies [17, 38]. Across all behaviors, the accelerometer-alone model generally outperformed the gyroscope-alone model, especially for static behaviors such as lying and standing. This trend suggests that accelerometer-derived features provide more distinguishable signals for postural states, whereas gyroscope signals tend to exhibit greater overlap between similar behaviors, a challenge also noted in livestock monitoring studies using ensemble classifiers [20]. Among the four behaviors, walking, which involves pronounced dynamic movement, posed the greatest classification challenge, with all sensor configurations demonstrating lower F1-scores and higher inter-cow variability. Despite these challenges, the combined-sensor model demonstrated improved accuracy and robustness, particularly notable in cows with low sensitivity in the gyroscope-only model. It is important to note that behaviors such as eating and walking were relatively underrepresented in the dataset. To mitigate potential classification bias arising from this class imbalance, weighted class penalties were applied during model training. This approach ensured that less frequent behaviors contributed proportionately to the learning process, thereby enhancing the model\u0026rsquo;s sensitivity to underrepresented behaviors. These findings underscore the capacity of Random Forest classifiers to exploit multidimensional, nonlinear features and effectively integrate heterogeneous sensor modalities for reliable behavior recognition across diverse individuals. The algorithm\u0026rsquo;s ensemble-based architecture enables it to capture nuanced, behavior-specific patterns within a high-dimensional feature space, reinforcing its suitability for deployment in precision livestock monitoring systems [39].\u003c/p\u003e\n\u003cp\u003e3.\u0026nbsp; \u0026nbsp;Contextualization within Prior Literature\u003c/p\u003e\n\u003cp\u003eThe findings of this study are broadly consistent with prior research that has employed accelerometer-only or gyroscope-only data for classifying cattle behavior, yet they extend the literature by offering a more nuanced axis-specific and individual-level analysis. Previous studies using accelerometer data alone have demonstrated acceptable classification performance for postural behaviors, particularly for lying and standing, due to the stability and directionality of gravitational acceleration [10]. Similarly, gyroscope-based models have shown advantages in capturing the rotational aspects of movement, especially during walking or transitional movements [17]. The present study advances beyond previous sensor-level approaches by decomposing axis-specific signal patterns and illustrating their behavioral relevance in a structured, cow-specific context. Unlike many earlier studies that aggregated data across animals, this investigation employed modeling at the individual-cow level, revealing behavior-specific variability in signal characteristics and classification performance, particularly in eating and walking. These findings suggest that cow-specific locomotor patterns may influence model performance, an aspect often underexplored in pooled analyses. Moreover, the comparative evaluation of sensor fusion demonstrated additive benefits, confirming that combining accelerometer and gyroscope data enhances model robustness across behavior types consistent with more recent multi-sensor livestock monitoring studies\u0026nbsp;[16, 36]. Some divergences from earlier findings may be attributable to differences in sensor placement, window length, sampling rate, or behavioral labeling strategies.\u003c/p\u003e\n\u003cp\u003e4.\u0026nbsp; \u0026nbsp;Strengths, Limitations, Application, and Future Directions\u003c/p\u003e\n\u003cp\u003eThis study presents several methodological strengths that enhance its scientific rigor and practical relevance. By conducting behavioral classification at the level of individual cows, the analysis captures inter-animal variability that is often overlooked in pooled modeling approaches. This individualized framework allows for a more precise understanding of cow-specific movement signatures and behavior expressions. Additionally, the structured axis-level evaluation of both accelerometer and gyroscope signals provides granular insight into the contribution of each sensor dimension, advancing the field\u0026rsquo;s understanding of movement ecology in dairy cattle. The use of Random Forest, a highly interpretable machine learning algorithm, further supports transparency and practical applicability through feature importance ranking, which aids in refining sensor deployment strategies and model development pipelines. Random Forest was selected due to its robustness to overfitting, capacity to manage high-dimensional and imbalanced datasets, and minimal assumptions about data distribution [40]. Its built-in feature importance metrics also support model interpretation, which is especially useful for sensor signal classification in animal behavior studies [20, 41].\u003c/p\u003e\n\u003cp\u003eNevertheless, certain limitations should be acknowledged. The relatively small sample size which was limited to seven heifers housed under consistent conditions may constrain the generalizability of findings to other dairy populations or production systems. However, the depth, duration, and resolution of the data collected (90 days of continuous sensing and over 780,000 labeled observations) provide strong analytical power for within-cow modeling. Although behavior annotation was performed manually, it followed a robust dual-observer protocol with high agreement (\u0026kappa; = 0.84), supported by joint training and consensus resolution. This enhances label reliability, even though manual annotation is time-consuming and may limit scalability in larger herds.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe scope of labeled behaviors was restricted to four clearly defined categories including lying, standing, eating, and walking while others such as ruminating or drinking were excluded due to low frequency or substantial behavioral overlap. This trade-off improved classification clarity but may limit behavioral comprehensiveness. Sensor placement was limited to the neck, which aligns with commercial practices but may reduce sensitivity to limb movement or asymmetric postures.\u003c/p\u003e\n\u003cp\u003eAn additional limitation involves the unequal distribution of behavioral data across individual cows, with some contributing disproportionately to specific behaviors such as walking and eating. This imbalance may introduce bias in model training and affect generalizability across individuals. However, this study addressed the issue through class-weighted modeling to ensure the representation of less frequent behaviors and by implementing cow-level analysis to preserve inter-individual variability. Class-weighted training has been validated as an effective strategy for handling imbalanced livestock behavior data, improving recognition of minority classes [42]. Furthermore, individual-cow modeling has been emphasized as a crucial technique for capturing heterogeneous behavioral expression in dairy cattle, which pooled models often overlook [13, 43]. These findings support the use of individualized modeling pipelines for precision livestock systems, particularly when behavior frequency and movement patterns differ markedly between animals.\u003c/p\u003e\n\u003cp\u003eFor practical implementation in commercial dairy systems, individualized behavior classification models must be supported by scalable calibration and rigorous validation under real-world conditions. The proposed models can be embedded in edge computing platforms integrated with wearable sensor units, enabling real-time feature extraction and behavior classification directly on the animal. This setup reduces reliance on continuous wireless transmission and supports low-latency monitoring. Data aggregation and advanced analytics can be conducted on centralized farm servers or cloud platforms, where long-term trends in individual behavior can inform herd-level decision-making.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo ensure adaptability to animal-specific behavior patterns, which is critical for the early detection of health and welfare deviations, future applications should incorporate flexible calibration workflows. These workflows could involve brief training periods using labeled behavioral data, such as short-term video annotations, to personalize models for newly introduced animals or production environments. Additionally, user interfaces should be developed to integrate model outputs into existing farm management software, thereby enabling automated alerts and decision-support tools for health surveillance, nutritional adjustments, and welfare monitoring.\u003c/p\u003e\n\u003cp\u003eIn addition to calibration, comprehensive external validation of the best-performing models across diverse environmental and management conditions is essential. This requirement is particularly relevant in regions with varied \u0026nbsp;environmental and management conditions, such as tropical areas where dairy operations often include loose housing, tie-stall barns, and pasture-based systems. Each of these settings presents distinct challenges related to behavioral expression, sensor signal quality, and variability in daily routines. Validation in such contexts is necessary not only to evaluate model robustness and adaptability, but also to refine behavioral thresholds, calibration strategies, and sensor configurations suited to heterogeneous farm environments. Rather than serving solely as a final performance check, validation should be considered an integral part of the model development process, contributing to the generalizability, operational reliability, and practical deployment of individualized behavior monitoring systems in precision livestock farming.\u003c/p\u003e\n\u003cp\u003eFuture research should aim to build on these findings by incorporating larger and more heterogeneous cow populations across multiple farms and management systems. Validation under commercial conditions will be essential to ensure the robustness and scalability of the proposed models. Furthermore, the integration of deep learning \u0026nbsp;and architectures, particularly those designed for temporal or sequential data (e.g., convolutional or recurrent neural networks), may enhance classification accuracy by learning context-aware movement patterns across time windows [44-46]. Finally, extending behavioral coverage to include overlapping or complex behaviors, such as ruminating, drinking, or social interaction, would increase the ecological relevance of automated monitoring systems and support broader applications in precision livestock management.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study demonstrated that integrating accelerometer and gyroscope data improves the accuracy and robustness of dairy cow behavior classification, particularly for static behaviors such as lying and standing. Axis-specific analysis revealed that both translational and rotational features contributed significantly to behavior differentiation, while individual-level modeling captured cow-specific variation in movement patterns. The Random Forest classifier performed reliably across behaviors and individuals by utilizing high-dimensional, multimodal features. Despite limitations related to sample size and herd homogeneity, this work establishes a strong foundation for sensor-based behavior monitoring in precision livestock systems. Future research should aim to validate these findings across diverse herds and environments, explore temporal deep learning architectures, and expand behavior recognition to include overlapping or complex activities.\u003c/p\u003e\n"},{"header":"Abbreviations","content":"\u003cp\u003eAccX, AccY, AccZ Accelerometer signals along X, Y, and Z axes\u003c/p\u003e\n\u003cp\u003eANOVA Analysis of Variance\u003c/p\u003e\n\u003cp\u003eCCTV Closed-Circuit Television\u003c/p\u003e\n\u003cp\u003eDBA Dynamic Body Acceleration\u003c/p\u003e\n\u003cp\u003eDBAX, DBAY, DBAZ Dynamic Body Acceleration on X, Y, and Z axes\u003c/p\u003e\n\u003cp\u003eFN False Negative\u003c/p\u003e\n\u003cp\u003eFP False Positive\u003c/p\u003e\n\u003cp\u003eGyroX, GyroY, GyroZ Gyroscope signals along X, Y, and Z axes\u003c/p\u003e\n\u003cp\u003eHSD Honest Significant Difference\u003c/p\u003e\n\u003cp\u003eIQR Interquartile Range\u003c/p\u003e\n\u003cp\u003eML Machine Learning\u003c/p\u003e\n\u003cp\u003eODBA Overall Dynamic Body Acceleration\u003c/p\u003e\n\u003cp\u003eRFE Recursive Feature Elimination\u003c/p\u003e\n\u003cp\u003eRF Random Forest\u003c/p\u003e\n\u003cp\u003eRMS Root Mean Square\u003c/p\u003e\n\u003cp\u003eSE Standard Error\u003c/p\u003e\n\u003cp\u003eSMA Signal Magnitude Area\u003c/p\u003e\n\u003cp\u003eSVM Signal Vector Magnitude\u003c/p\u003e\n\u003cp\u003eSVM_acc Signal Vector Magnitude from accelerometer\u003c/p\u003e\n\u003cp\u003eSVM_gyro Signal Vector Magnitude from gyroscope\u003c/p\u003e\n\u003cp\u003eTP True Positive\u003c/p\u003e\n\u003cp\u003eWi-Fi Wireless Fidelity\u003c/p\u003e\n\u003cp\u003eVeDBA Vectorial Dynamic Body Acceleration\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u0026middot; \u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Institutional Animal Care and Use Committee (IACUC) of the Faculty of Veterinary Science, Chulalongkorn University, Thailand (Protocol No. 2031047), in accordance with institutional regulations and the Ethical Principles and Guidelines for the Use of Animals for Scientific Purposes issued by the National Research Council of Thailand.\u003c/p\u003e\n\u003cp\u003e\u0026middot; \u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u0026middot; \u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe cow activity data generated and analyzed during this study is available upon reasonable request from the Research Unit of Data Innovation for Livestock, Department of Veterinary Medicine, Faculty of Veterinary Science, Chulalongkorn University, Bangkok, Thailand. To support reproducibility, a subset of the data comprising both raw and processed activity data is publicly available at the GitHub repository: https://github.com/pongsanun. In addition, all Python scripts used for data manipulation, feature engineering, visualization, and model development, as well as R scripts used for statistical analysis, are provided in the same repository.\u003c/p\u003e\n\u003cp\u003e\u0026middot; \u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests\u003c/p\u003e\n\u003cp\u003e\u0026middot; \u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was supported by the Research and Researchers for Industries (RRI) program (Contract No. PHD60I0084), the Program Management Unit for Competitiveness (PMUC) (Contract No. 1499287), and the Research Unit of Data Innovation for Livestock, Department of Veterinary Medicine, Faculty of Veterinary Science, Chulalongkorn University, Bangkok, Thailand.\u003c/p\u003e\n\u003cp\u003e\u0026middot; \u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePK and CI were responsible for the study design and overall conceptualization. AT, AL, and SC contributed to the design and development of the sensor prototypes and data transmission systems used for data collection. PK was responsible for prototype installation and system monitoring to ensure successful data acquisition. Behavioral data annotation was conducted by PK and AT. All data analyses were performed by PK. The manuscript was drafted by PK with critical revisions and editorial input from CI. All authors have read and approved the final version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u0026middot; \u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors would like to express their sincere gratitude to the Director of the Farm Animal Hospital, Faculty of Veterinary Science, Chulalongkorn University, Nakhon Pathom, Thailand, for granting permission to conduct the experiment. The authors also extend their heartfelt appreciation to the animal husbandry team and hospital staff for their continuous support and kind assistance throughout the experimental period.\u003c/p\u003e\n\u003cp\u003e\u0026middot; \u003cstrong\u003eAuthors\u0026apos; information (optional)\u003c/strong\u003e\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eQiao Y, Guo Y, Yu K, He D: C3D-ConvLSTM based cow behaviour classification using video data for precision livestock farming. Computers and Electronics in Agriculture 2022, 193:106650.\u003c/li\u003e\n\u003cli\u003eNorton T, Berckmans D: Developing precision livestock farming tools for precision dairy farming. Animal Frontiers 2017, 7(1):18-23.\u003c/li\u003e\n\u003cli\u003eHalachmi I, Guarino M: Editorial: Precision livestock farming: a \u0026lsquo;per animal\u0026rsquo; approach using advanced monitoring technologies. animal 2016, 10(9):1482-1483.\u003c/li\u003e\n\u003cli\u003eWalker SL, Smith RF, Routly JE, Jones DN, Morris MJ, Dobson H: Lameness, Activity Time-Budgets, and Estrus Expression in Dairy Cattle. Journal of Dairy Science 2008, 91(12):4552-4559.\u003c/li\u003e\n\u003cli\u003eWeigele HC, Gygax L, Steiner A, Wechsler B, Burla JB: Moderate lameness leads to marked behavioral changes in dairy cows. Journal of Dairy Science 2018, 101(3):2370-2382.\u003c/li\u003e\n\u003cli\u003ePereira GM, Sharpe KT, Heins BJ: Evaluation of the RumiWatch system as a benchmark to monitor feeding and locomotion behaviors of grazing dairy cows. Journal of Dairy Science 2021, 104(3):3736-3750.\u003c/li\u003e\n\u003cli\u003eRiaboff L, Shalloo L, Smeaton AF, Couvreur S, Madouasse A, Keane MT: Predicting livestock behaviour using accelerometers: A systematic review of processing techniques for ruminant behaviour prediction from raw accelerometer data. Computers and Electronics in Agriculture 2022, 192:106610.\u003c/li\u003e\n\u003cli\u003eBenaissa S, Tuyttens FAM, Plets D, Martens L, Vandaele L, Joseph W, Sonck B: Improved cattle behaviour monitoring by combining Ultra-Wideband location and accelerometer data. animal 2023, 17(4):100730.\u003c/li\u003e\n\u003cli\u003eRob\u0026eacute;rt B, White B, Renter D, Larson R: Evaluation of three-dimensional accelerometers to monitor and classify behavior patterns in cattle. Computers and Electronics in Agriculture 2009, 67:80-84.\u003c/li\u003e\n\u003cli\u003eMartiskainen P, J\u0026auml;rvinen M, Sk\u0026ouml;n J-P, Tiirikainen J, Kolehmainen M, Mononen J: Cow behaviour pattern recognition using a three-dimensional accelerometer and support vector machines. Applied Animal Behaviour Science 2009, 119(1):32-38.\u003c/li\u003e\n\u003cli\u003ePereira GM, Heins BJ, Endres MI: Technical note: Validation of an ear-tag accelerometer sensor to determine rumination, eating, and activity behaviors of grazing dairy cattle. Journal of Dairy Science 2018, 101(3):2492-2495.\u003c/li\u003e\n\u003cli\u003eRussel NS, Selvaraj A: Decoding cow behavior patterns from accelerometer data using deep learning. Journal of Veterinary Behavior 2024, 74:68-78.\u003c/li\u003e\n\u003cli\u003eWilliams ML, Mac Parthal\u0026aacute;in N, Brewer P, James WPJ, Rose MT: A novel behavioral model of the pasture-based dairy cow from GPS data using data mining and machine learning techniques. Journal of Dairy Science 2016, 99(3):2063-2075.\u003c/li\u003e\n\u003cli\u003eAnzai H, Hirata M: Individual Monitoring of Behavior to Enhance Productivity and Welfare of Animals in Small-Scale Intensive Cattle Grazing Systems. Frontiers in Sustainable Food Systems 2021, Volume 5 - 2021.\u003c/li\u003e\n\u003cli\u003eSinghal G, Choudhary P, Abhishek V, Sweety S, Subramanian S, Goel N: Cattle Collar: An End-to-End Multi-Model Framework for Cattle Monitoring. In: 2022 IEEE 5th International Conference on Multimedia Information Processing and Retrieval (MIPR): 2-4 Aug. 2022 2022; 2022: 401-407.\u003c/li\u003e\n\u003cli\u003eBenaissa S, Tuyttens FAM, Plets D, de Pessemier T, Trogh J, Tanghe E, Martens L, Vandaele L, Van Nuffel A, Joseph W et al: On the use of on-cow accelerometers for the classification of behaviours in dairy barns. Research in Veterinary Science 2019, 125:425-433.\u003c/li\u003e\n\u003cli\u003eMladenova T, Valova I, Evstatiev B, Valov N, Varlyakov I, Markov T, Stoycheva S, Mondeshka L, Markov N: Evaluation of the Efficiency of Machine Learning Algorithms for Identification of Cattle Behavior Using Accelerometer and Gyroscope Data. In: AgriEngineering. vol. 6; 2024: 2179-2197.\u003c/li\u003e\n\u003cli\u003eObermeyer K, Kayser M: On-farm assessment of grazing behaviour of dairy cows in two pasture management systems by low-cost and reliable cowtrackers. Smart Agricultural Technology 2023, 6:100349.\u003c/li\u003e\n\u003cli\u003eWilliams M, Zhan Lai S: Classification of dairy cow excretory events using a tail-mounted accelerometer. Computers and Electronics in Agriculture 2022, 199:107187.\u003c/li\u003e\n\u003cli\u003eMartono NP, Sawado R, Nonaka I, Terada F, Ohwada H: Automated Cattle Behavior Classification Using Wearable Sensors and Machine Learning Approach. In: Knowledge Management and Acquisition for Intelligent Systems: 2023// 2023; Singapore: Springer Nature Singapore; 2023: 58-69.\u003c/li\u003e\n\u003cli\u003ePokydko M, Oliinyk O, Tymchenko V: MEMS Gyroscope Based on MPU-6050 Sensor and ATmega328 Microcontroller. In: 2024 IEEE 7th International Conference on Smart Technologies in Power Engineering and Electronics (STEE): 24-26 Sept. 2024; 2024: TT3.39.31-TT33.39.36.\u003c/li\u003e\n\u003cli\u003eDang TH, Dang NH, Tran VT, Chung WY: A LoRaWAN-Based Smart Sensor Tag for Cow Behavior Monitoring. In: 2022 IEEE Sensors: 30 Oct.-2 Nov. 2022; 2022: 1-4.\u003c/li\u003e\n\u003cli\u003eCatrett CC, Parsons IL, Dentinger JE, Norman DA, Webb SL, Stone AE, Street G, Karisch BB: PSII-12 Identifying behaviors and the \u0026lsquo;normal\u0026rsquo; daily ethogram using accelerometers on grazing animals. Journal of Animal Science 2021, 99(Supplement_3):319-320.\u003c/li\u003e\n\u003cli\u003eLi K, Fan D, Wu H, Zhao A: A new dataset for video-based cow behavior recognition. Scientific Reports 2024, 14(1):18702.\u003c/li\u003e\n\u003cli\u003eRiaboff L, Aubin S, B\u0026eacute;d\u0026egrave;re N, Couvreur S, Madouasse A, Goumand E, Chauvin A, Plantier G: Evaluation of pre-processing methods for the prediction of cattle behaviour from accelerometer data. Computers and Electronics in Agriculture 2019, 165:104961.\u003c/li\u003e\n\u003cli\u003eWilliams ML, James WP, Rose MT: Variable segmentation and ensemble classifiers for predicting dairy cow behaviour. Biosystems Engineering 2019, 178:156-167.\u003c/li\u003e\n\u003cli\u003eNogoy KM, Chon S-i, Park J-h, Sivamani S, Lee D-H, Choi SH: High Precision Classification of Resting and Eating Behaviors of Cattle by Using a Collar-Fitted Triaxial Accelerometer Sensor. In: Sensors. vol. 22; 2022.\u003c/li\u003e\n\u003cli\u003ePavlovic D, Czerkawski M, Davison C, Marko O, Michie C, Atkinson R, Crnojevic V, Andonovic I, Rajovic V, Kvascev G et al: Behavioural Classification of Cattle Using Neck-Mounted Accelerometer-Equipped Collars. In: Sensors. vol. 22; 2022.\u003c/li\u003e\n\u003cli\u003eArablouei R, Currie L, Kusy B, Ingham A, Greenwood PL, Bishop-Hurley G: In-situ classification of cattle behavior using accelerometry data. Computers and Electronics in Agriculture 2021, 183:106045.\u003c/li\u003e\n\u003cli\u003eUenishi S, Oishi K, Kojima T, Kitajima K, Yasunaka Y, Sakai K, Sonoda Y, Kumagai H, Hirooka H: A novel accelerometry approach combining information on classified behaviors and quantified physical activity for assessing health status of cattle: a preliminary study. Applied Animal Behaviour Science 2021, 235:105220.\u003c/li\u003e\n\u003cli\u003eYin Z, Liu L, Liu L, Zhang J, Wang Y: Dynamical recursive feature elimination technique for neurophysiological signal-based emotion recognition. Cognition, Technology \u0026amp; Work 2017, 19(4):667-685.\u003c/li\u003e\n\u003cli\u003eShekar BH, Dagnew G: Grid Search-Based Hyperparameter Tuning and Classification of Microarray Cancer Data. In: 2019 Second International Conference on Advanced Computational and Communication Paradigms (ICACCP): 25-28 Feb. 2019; 2019: 1-8.\u003c/li\u003e\n\u003cli\u003eJim\u0026eacute;nez A, Bautista F, Galina CS, Romero JJ, Rubio I: Behavioral Characteristics of Bos indicus Cattle after a Superovulatory Treatment Compared to Cows Synchronized for Estrus. Asian-Australas J Anim Sci 2011, 24(10):1365-1371.\u003c/li\u003e\n\u003cli\u003eLozada CC, Park RM, Daigle CL: Evaluating accurate and efficient sampling strategies designed to measure social behavior and brush use in drylot housed cattle. PLOS ONE 2023, 18(1):e0278233.\u003c/li\u003e\n\u003cli\u003eHirata M, Nakayama Y, Tobisa M: Interindividual variability in feeding station behavior in cattle: A preliminary study. Grassland Science 2010, 56(2):108-115.\u003c/li\u003e\n\u003cli\u003eTamura T, Okubo Y, Deguchi Y, Koshikawa S, Takahashi M, Chida Y, Okada K: Dairy cattle behavior classifications based on decision tree learning using 3-axis neck-mounted accelerometers. Animal Science Journal 2019, 90(4):589-596.\u003c/li\u003e\n\u003cli\u003eSprinkle JE, Sagers JK, Hall JB, Ellison MJ, Yelich JV, Brennan JR, Taylor JB, Lamb JB: Predicting Cattle Grazing Behavior on Rangeland using Accelerometers. Rangeland Ecology \u0026amp; Management 2021, 76:157-170.\u003c/li\u003e\n\u003cli\u003eKleanthous N, Hussain A, Mason A, Sneddon J: Data Science Approaches for the Analysis of Animal Behaviours. In: Intelligent Computing Methodologies: 2019// 2019; Cham: Springer International Publishing; 2019: 411-422.\u003c/li\u003e\n\u003cli\u003eKleanthous N, Hussain A, Khan W, Sneddon J, Mason A: Feature Extraction and Random Forest to Identify Sheep Behavior from Accelerometer Data. In: Intelligent Computing Methodologies: 2020// 2020; Cham: Springer International Publishing; 2020: 408-419.\u003c/li\u003e\n\u003cli\u003eBreiman L: Random Forests. Machine Learning 2001, 45(1):5-32.\u003c/li\u003e\n\u003cli\u003eIbrahim T, Isaac KB, Francis B, Lule E, Hellen N, Chongomweru H, Marvin G: Interpretable Machine Learning Techniques for Predictive Cattle Behavior Monitoring. In: 2024 2nd International Conference on Sustainable Computing and Smart Systems (ICSCSS): 10-12 July 2024; 2024: 1219-1224.\u003c/li\u003e\n\u003cli\u003eDaker M, Elsayaad F, Atia A: The Classification Of Cattle Behaviors Using Deep Learning. In: 2024 6th International Conference on Computing and Informatics (ICCI): 6-7 March 2024 2024; 2024: 28-33.\u003c/li\u003e\n\u003cli\u003eArcidiacono C, Porto SMC, Mancino M, Cascone G: Development of a threshold-based classifier for real-time recognition of cow feeding and standing behavioural activities from accelerometer data. Computers and Electronics in Agriculture 2017, 134:124-134.\u003c/li\u003e\n\u003cli\u003eHosseininoorbin S, Layeghy S, Kusy B, Jurdak R, Bishop-Hurley GJ, Greenwood PL, Portmann M: Deep learning-based cattle behaviour classification using joint time-frequency data representation. Computers and Electronics in Agriculture 2021, 187:106241.\u003c/li\u003e\n\u003cli\u003eGao G, Wang C, Wang J, Lv Y, Li Q, Ma Y, Zhang X, Li Z, Chen G: CNN-Bi-LSTM: A Complex Environment-Oriented Cattle Behavior Classification Network Based on the Fusion of CNN and Bi-LSTM. In: Sensors. vol. 23; 2023.\u003c/li\u003e\n\u003cli\u003ePeng Y, Kondo N, Fujiura T, Suzuki T, Wulandari, Yoshioka H, Itoyama E: Classification of multiple cattle behavior patterns using a recurrent neural network with long short-term memory and inertial measurement units. Computers and Electronics in Agriculture 2019, 157:247-253.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Table 2","content":"\u003cp\u003eTable 2 is available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-veterinary-research","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [BMC Veterinary Research](http://bmcvetres.biomedcentral.com/)","snPcode":"12917","submissionUrl":"https://submission.nature.com/new-submission/12917/3?","title":"BMC Veterinary Research","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Dairy cow behavior, Behavior classification, Precision livestock farming, Random Forest machine learning, Sensor fusion, Tri-axial accelerometer and gyroscope","lastPublishedDoi":"10.21203/rs.3.rs-6682405/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6682405/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eBackground: Automated behavior monitoring is increasingly important in precision dairy farming, supporting early disease detection, welfare assessment, and productivity optimization. Although accelerometers effectively detect postural changes, they have limited capacity to capture rotational or transitional movements. Gyroscopes provide complementary angular velocity data that may enhance classification of complex behaviors. However, their combined use remains underexplored, particularly at the individual cow level. This study aims to evaluate the performance of accelerometer, gyroscope, and combined sensors models for classifying four key cow behaviors: lying, standing, eating, and walking at the individual animal level.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eResults: Over 780,000 labeled observations were obtained from seven dairy cows monitored over 90 days. Lying behavior consistently produced low, stable signals across all axes of accelerometer and gyroscope, while eating showed the greatest variability, particularly along the X and Y axes. Significant axis-specific and behavior-specific differences were observed (p \u0026lt; 0.05), with GyroY and GyroZ capturing the highest rotational activity during eating and walking. Signal vector magnitudes effectively distinguished behaviors, with lying showing the lowest values and eating the highest. Random Forest models combining accelerometer and gyroscope data consistently outperformed single-sensor approaches, particularly for classifying lying and standing behaviors. Although eating and walking exhibited lower sensitivity, sensor fusion improved classification robustness across individuals.\u003c/p\u003e\n\u003cp\u003eConclusion: The integration of accelerometer and gyroscope data enhanced classification accuracy, particularly for static behaviors. Axis-specific signal patterns and individualized modeling revealed critical insights into behavior differentiation and cow-specific variability. These findings support the development of scalable, sensor-based monitoring systems tailored to precision livestock management.\u003c/p\u003e","manuscriptTitle":"Classification of Individual Dairy Cow Behaviors Using Accelerometer, Gyroscope, and Integrated Sensor Models","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-29 18:10:22","doi":"10.21203/rs.3.rs-6682405/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-08-18T05:05:31+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-15T06:22:55+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-12T20:04:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"2804851322810348075249004380862120015","date":"2025-08-12T05:32:00+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"290467906427956068747551247040286483950","date":"2025-08-04T18:08:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"98227330895092810088508343757038333702","date":"2025-07-22T07:15:21+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-05-28T07:34:54+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-05-21T06:48:35+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-05-21T06:48:02+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Veterinary Research","date":"2025-05-16T16:26:24+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-veterinary-research","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [BMC Veterinary Research](http://bmcvetres.biomedcentral.com/)","snPcode":"12917","submissionUrl":"https://submission.nature.com/new-submission/12917/3?","title":"BMC Veterinary Research","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"fcf40a1e-1c2d-4712-ab39-af8b0586ebff","owner":[],"postedDate":"May 29th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-11-24T16:05:43+00:00","versionOfRecord":{"articleIdentity":"rs-6682405","link":"https://doi.org/10.1186/s12917-025-05092-1","journal":{"identity":"bmc-veterinary-research","isVorOnly":false,"title":"BMC Veterinary Research"},"publishedOn":"2025-11-20 15:57:52","publishedOnDateReadable":"November 20th, 2025"},"versionCreatedAt":"2025-05-29 18:10:22","video":"","vorDoi":"10.1186/s12917-025-05092-1","vorDoiUrl":"https://doi.org/10.1186/s12917-025-05092-1","workflowStages":[]},"version":"v1","identity":"rs-6682405","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6682405","identity":"rs-6682405","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}