A Novel Multivariate Time Series Dataset of Outdoor Sport Activities

A Novel Multivariate Time Series Dataset of Outdoor Sport Activities | 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 A Novel Multivariate Time Series Dataset of Outdoor Sport Activities Jarno Olavi Matarmaa This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4593851/v2 This work is licensed under a CC BY 4.0 License Status: Under Review Version 2 posted 8 You are reading this latest preprint version Show more versions Abstract This study introduces a novel multivariate time series dataset of 228 outdoor sport activities recorded by individual non-competitive athlete in uncontrolled environments. The dataset includes three features: Heart Rate, Speed, and Altitude, and covers five sport categories: walking, running, skiing, roller-skiing, and biking. The data was collected using two types of Garmin sport watches. The original dataset was carefully pre-processed using typical data cleansing methods such as gaps filling, and value format transformations. Furthermore, activity filtering was implemented for missing sensor value data and using domain knowledge of sport categories. Full length sequences, varying from 10 minutes to several hours, were split into equal length segments, approximately 1 minute. To address the small number of instances data was augmented using several consecutive segments from the same activity. However, only a small part of the whole original data was used as a computational cost–information gain tradeoff. Three-dimensional dataset is divided into three parts, each dimension to its own comma separated value (CSV) file. The dataset aims to provide a unique resource for researchers and practitioners in the field of sports science, human performance analysis, and activity recognition. It aims to complement the very limited or non-existent publicly available sport activity datasets. multivariate time series outdoor sport sport exercises sport dataset Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 1. Introduction The research domain of the newly introduced novel sport activity dataset is originally motivated and related, although not limited to the field of Human Activity Recognition (HAR). A field that has witnessed extensive exploration over the last two decades, propelled by the surging popularity and advancement of activity bracelets, smartwatches, and smartphones equipped with inertial sensors for data collection [ 1 ]. The evolution of deep learning and machine learning algorithms has enabled the integration of HAR into various domains, spanning sports, health, and well-being applications. HAR's overarching goal is to recognize human activities in both controlled and uncontrolled settings. Lara and Labrador (2013) contributed a comprehensive overview to the field of HAR studies, specifically addressing personalized sport activity recognition [ 2 ]. Despite subsequent algorithmic developments, their study remains influential, notably in highlighting the ongoing debate on activity recognition model design. Some studies argue for the construction of specific recognition models for individual differences in age, gender, and weight, necessitating system retraining for each new user [ 3 ]. Conversely, other studies emphasize the need for a monolithic recognition model adaptable across diverse users [ 4 ]. This debate has given rise to two types of analyses for evaluating activity recognition systems: subject-dependent and subject-independent evaluations [ 5 ]. In subject-dependent evaluations, a classifier is trained and evaluated for each individual, and the average accuracy across all subjects is computed. In subject-independent evaluations, a single classifier is constructed for all individuals using cross-validation or leave-one-individual-out analysis. Lara and Labrador highlighted practical challenges in retraining systems for each new user, particularly in cases with numerous activities, undesirable activities, or uncooperative subjects. Recognizing that individuals, especially across age groups, may exhibit different activity patterns, they proposed addressing the dichotomy of monolithic and personalized models by creating groups of users with similar characteristics [ 2 ]. However, the practical implementation of this solution is often constrained by the availability of suitable datasets in HAR studies. Despite achieving impressive accuracies of up to 99% in controlled experiments, concerns persist regarding the realism and environmental variability of data collection settings. Notably, significant accuracy drops have been observed since the beginning of HAR studies in late 1990s when transitioning from controlled laboratory experiments to uncontrolled, non-laboratory natural environments [ 6 ]. Consequently, there is a growing emphasis on conducting studies in more challenging datasets to ensure comprehensive consideration of all environmental variables. For HAR studies, wherein inertial sensor data of three-dimensional-accelerometers are usually applied there are quite numerous notable datasets [ 1 , 7 – 10 ], some of them easily accessible as they are indexed by search engines. Some of the most common datasets are listed in Table 1 . However, for example, sport activity recognition does not need to be limited to inertial sensors. Rather, the hypothesis could be set in the following way: can we recognize sports using only physiological and environmental data types, instead of inertial? For instance, what is the response of heart rate when there are environmental changes like derivative of altitude, or even base level of altitude? Not only between different sports, but also the effect of altitude among same category of activities. What about the relationship of heart rate and speed: how much heart rate will increase if the speed increases by 10%? For different type of sports, the response of the heart rate for the changes in speed might significantly differ. This type of dataset could allow us to conduct remarkably another level of studies, although it is limited to outdoor activities due to technological limitations of GPS based speed and air pressure-based altitude measurements. Table 1 Public multivariate time series datasets for Human and Sport Activity Recognition. Dataset Year, Study Data Type / Tasks Classes Instances Sensors UCI HAR 1 2012, [ 10 ],[ 11 ] Classification, Clustering 6 10299 2 x 3-axial UCI HAR 2 2016, [ 12 ] Classification, Clustering 6 5744 2 x 3-axial HAR70+ 2023, [ 8 ] Classification 8 2 259 597 2 x 3-axial Daily and Sport Activities 2010, [ 7 ] Classification, Clustering 19 9120 9 x 3-axial BasicMotions 2000s, [ 13 ] Classification (Education) 4 80 2 x 3-axial UniMiB SHAR 2016, [ 9 ] Classification 9 11 771 1 x 3-axial To address these critical considerations, this study introduces and publishes a new and unique dataset recorded in a genuine end user environment, and permission for research use has been granted later. Multivariate time series data, which includes multiple variables measured over time, can provide valuable insights into the physiological and environmental factors that influence athletic performance. However, publicly available datasets that capture the complexity of outdoor sports in uncontrolled or controlled environments are either nonexistent or limited. A novel dataset was created that includes 228 outdoor sport activities across five categories: walking, running, skiing, roller-skiing, and biking. The dataset was collected by a non-competitive male athlete using Garmin Forerunner 920XT and Vivosport devices, which recorded heart rate, speed, and altitude data. These devices were selected for their compatibility within the same sport activity tracking ecosystem and their use of different technologies for heart rate measurement, with the Vivosport employing an optical sensor. This choice helps mitigate device-specific bias in data collection. The dataset offers a unique opportunity for researchers and practitioners to explore the relationships between physiological and environmental factors in outdoor sports. The objective is to present a thorough overview of the dataset’s structure, the preprocessing and segmentation processes employed, and the substantial potential it holds for multifaceted research in activity recognition, endurance training, and performance analytics. Detailed representation of the dataset and its five distinct outdoor sports activities are provided. Each activity is recorded under natural, uncontrolled conditions, offering a rare glimpse into the dynamics of sports performance outside laboratory settings. This dataset is uniquely positioned to facilitate nuanced research into how environmental factors influence sports performance and can help in the development of adaptive training programs tailored to individual physiological responses. However, it is important to recognize that environmental factors associated with certain sports, such as skiing, are significantly influenced by individual differences. Skiing can be performed under various conditions and environments, depending on the athlete's geolocation and personal preferences. This variability does not diminish the value of a one-man study; however, it should be considered when applying the dataset to different contexts or applications. Furthermore, the longitudinal nature of the data collection provides a rich canvas to explore seasonal variations and trends over time, making it an invaluable resource for long-term studies in sports and health disciplines. Moreover, by making this dataset publicly available, it encourages open scientific inquiry and collaborative research efforts that can lead to breakthroughs in predictive modelling and real-time monitoring in sports and health sectors. 2. Methodology Data collection The data collection of activities was recorded over a period of 16 months by a non-competitive male athlete, employing wearable devices equipped with multiple sensors. Each activity was recorded under natural, uncontrolled environmental conditions to capture realistic performance data. The athlete’s consistent participation over this extended period allowed for the collection of a comprehensive set of data across different seasons and varying weather conditions and environments. The data recording devices of the dataset are Garmin Forerunner 920XT with HRM-Run sensor as shown in the figure series Fig. 1, and Garmin Vivosport activity tracker which has optical wrist-based heart rate measurement technology. Vivosport was used to record a quantitative minority of sports, the majority of them walking activities. The study followed a generic data acquisition architecture for HAR system, as depicted in the study [ 2 ]. This architecture included wearable sensors, as shown in Fig. 1, integration devices such as laptops or smartphones, and communication protocols like TCP/IP to transfer data to a local computer or remote cloud storage. Recording the activities using a sport watch is a straightforward process. When the user starts recording the activity, a proper predetermined sport profile is manually selected from the device. Sport profile selection sets the label for sport activity, and therefore this action is prone to misclassification due to human nature. When completing the activity, the user manually stops the recording and saves it to the device’s memory. The next step after saving activity to the memory of sport watch is to send it to a smartphone application, the process called synchronization of devices. Another option is to send activity via Wi-Fi connection to the cloud service which was available in the Garmin FR-920XT used in data recording. The challenging part of the data collection appears while accessing the data from the cloud service, in this case connect.garmin.com . These commercial service providers follows the data handling policies set by governments such as policy of data portability in GDPR [ 14 ], providing full rights of the data to the data owner, but practically they often offer very limited tools or ways for data export. Therefore, a massive data processing was required including conversions from a complex TCX (Training Center XML) file format to a CSV (Comma Separated Value) format. The specific data conversion algorithm for this purpose was developed, after which data pre-processing using traditional Python data frames was allowed. Data dimensions The sport tracking devices enabled a maximum sampling rate with a one-second time interval. The dataset encapsulates three primary dimensions measured using distinct sensors embedded in the wearable devices: heart rate, geolocation, and barometric pressure. These sensors provided continuous measurement streams that were then transformed into interpretable metrics: Heart Rate Captured directly from a heart rate sensor, reflecting the athlete's physiological response to the activity. The original values are given in beats per minute (bpm), with an average range typically observed from 40 to 220 bpm, depending on the intensity of the activity. Speed Derived from the geolocation data, calculated to show the pace at which the athlete travelled over ground. Measured in meters per second (m/s), with values ranging from 0 (when stationary) up to 20 m/s, particularly noticeable in faster-paced activities like biking. Altitude Obtained from the barometric pressure sensor, adjusted to reflect elevation changes during the activities. Altitude values are recorded in meters (m), reflecting the elevation gains or losses during activities, which could help understanding of terrain impacts on performance. These dimensions were specifically chosen for their relevance to assessing athletic performance and their ability to provide insights into the physical demands of different outdoor sport activities. Heart rate could represent an engine that will provide some speed, whereas their correlation is highly affected by the value and changes in altitude. The stress on the uphill is considerable higher than on the downhill leading to an increased heart rate, if speed remains the same, according to the laws of physics and energy. And the heart rate increases as the speed increases on flat ground. However, in biking activity heart rate usually decreases while speed is increasing on the downhill. But it is also possible, that person amplifies acceleration on the downhill using physical effort, leading to an aggressive speed increasement. This forms a rather fascinating sensor data combination. Preprocessing and cleaning The raw data collected underwent meticulous preprocessing, cleaning, and filtering to ensure high quality and usability. Data Cleaning Activities with missing sensor data or recordings of insufficient length were excluded from the dataset. This step was crucial to maintaining the integrity of the dataset, particularly given its relatively small size. Segmentation The data was segmented into 60 seconds intervals. This segmentation process began at the 100-second mark of each recorded activity to avoid initial anomalies and stabilize sensor readings. The choice of starting at 100 seconds and using 60 seconds segment length was based on a combination of visual data analysis and domain-specific knowledge, optimizing for both computational efficiency and classification accuracy of preliminary tests conducted in the dataset. Standardization All data was standardized to ensure uniformity across different measures, facilitating more accurate analysis and comparison. Features were standardized by centering them around the mean and scaling to unit variance. The standard score of a sample x was calculated as $$\:z=\frac{x-\mu\:}{\sigma\:}$$ where µ is the mean of the signal, and σ is the standard deviation of the signal attribute. Data Augmentation To address the dataset's limited size and enhance the robustness of subsequent analyses, data augmentation was performed by selecting five consecutive one-minute segments from the same activity. This approach effectively increased the volume of data for each activity type without compromising the natural variability inherent in the athlete's performance. The applied data cleansing function processed time series data for activity classification by first loading specified entries, CSV files into data frames, then configuring their indices as datetime objects to facilitate time-based analysis. Relevant features, heart rate, speed, and altitude, were filtered from each data frame. Data cleaning involves replacing placeholder strings with numerical missing values (NaN), converting all data to a numeric format, and addressing missing values through a series of methods: linear interpolation is used first to fill gaps based on surrounding data; remaining gaps are filled by carrying forward the last known value; and any persistently missing values are set to zero. This thorough cleaning ensures the data's continuity and usability for classification algorithms. The following cleaning steps were applied: Replacing placeholders : Converting 'None' strings to NaN to standardize missing value representation. Ensuring numeric data : Transforming all feature values to numeric types, handling any conversion errors gracefully. Interpolating missing values : Filling NaN values by linear interpolation using forward-backward filling method to maintain data continuity. Zero filling : Setting any remaining missing values to zero, ensuring no gaps in the data sequence. The results of pre-processing and filtering are depicted in the Table 2 . The final pre-processed dataset comprises 1140 segments, each 60 seconds in length, across the three dimensions of heart rate, speed, and altitude. Each dimension is stored in separate CSV files to maintain clarity and ease of access for analysis purposes. The segmentation of the original processed sequences was implemented using that clean and filtered data these segments are the end product of this study, the public dataset of sport activities in five categories. Table 2 Sequence/Signal length statistics before and after applying data filtering. Feature Original data Clean data Change % count 280.000 228.000 -18.57 mean length (s) 4041.621 4584.307 + 11.86 standard deviation 3077.461 2793.751 -10.16 min length (s) 0.000 524.000 + 524.00 25% 1485.750 2785.500 + 87.48 50% 3849.000 4437.000 + 15.28 75% 5647.250 5910.500 + 4.66 max length (s) 19781.000 19781.000 0.00 3. Results Data structure and format The data is stored in four separate CSV files for ease of use and compatibility with most data analysis tools. Each CSV file is named according to the feature type and dimensions, except meta data with label names and device info. For instance, files are named in the format: FEATURE-DATA_std_NxM.csv (e.g., HR-DATA_std_1140x69.csv). So called metadata with sport category labels and recording device names were stored in the fourth file in corresponding indexes. The order of instances in these four files must not be confused when processing them. The dataset is structured as a collection of univariate time series, each corresponding to a different dimension of original multivariate dataset. In other words, each time series segment of 1 minute is structured into three files representing the three recorded dimensions: heart rate, speed, and altitude. Each CSV file contains time series data in a transposed matrix format, where columns represent sequential time points and rows correspond to the different sport activities. This transposed format allows us to use traditional data structure for classification tasks which is generally used as a valid input format in the machine learning library algorithms (e.g. Scikit-Learn). Thus, corresponding indexes in each feature file represents the single activity. There are several ways for structuring multivariate three-dimensional data in two-dimensional space, but the described one was used as most convenient structure as the dataset is designed to be used specifically in classification tasks, among others. Data characteristics and features The dataset consists of 1140 segments, with each segment in length of 60 seconds. The dataset encompasses a total of 228 recorded activities, divided into categories and recording devices according to the Fig. 2. Data samples can have many types of patterns and value levels and their combinations. For instance, as we can see from the Fig. 3 , some random three-dimensional segment A (Activity A) has lower-level speed value and smaller variance, whereas in segment B (Activity B) speed value is fluctuating noticeably more in sixty seconds time window. This could be potential indicator of diverse types of sport, especially when considering heart rate value at the same time window. Altitude with remarkably similar decreasing trend probably indicates that activities are performed in the same environment. That could be one of the ways to interpret and investigate the data. We use violin plots to observe the most obvious value range characteristics of the sports as seen in Figs. 4 , 5 , and 6 . Violin plots depict effectively how the values are distributed providing more information compared to traditional boxplot. We can detect different kinds of shapes for each category and thus suppose sport specific behaviour for heart rate, speed, and altitude features. To further illustrate the distinctiveness and characteristics of the data between features and categories we could use very descriptive mean value segments as in Fig. 7 . It is likely the easiest way of understanding the nature of the segmented data. To mention a few, for instance, biking is clearly the fastest sport among all the others and walking the slowest. Also, biking segments has very low hear rate value emphasizing the effectiveness of the sport, as a higher heart rate during the sport activity leads often more rapidly to physical fatigue. Noticeable observation is that altitude feature has smaller gaps among sports which should be considered when conducting any machine learning task in the data. Kernel density estimations Given that the dataset is three-dimensional, with altitude lacking a clear relationship with speed and heart rate, the latter two attributes are chosen for visualization due to their notable positive correlation. Figure 8 illustrates the distribution of speed-heart rate value points across categories. Notably, Skiing and R-Skiing exhibit similar patterns, while other categories demonstrate more discriminative characteristics. When kernel density graphics of categories are amalgamated into a single figure, intersection areas will become more apparent. Density values are normalized by category to account for an uneven distribution of instances. t-SNE Lastly, we applied t-distributed stochastic neighbor embedding algorithm for clustering the data. T-SNE is a statistical method for visualizing multi-dimensional data by giving each datapoint a location in a two or three-dimensional map. [ 15 , 16 ] While t-SNE plots often appear to display clusters, the visual clusters can be significantly influenced by the chosen parameterization. Therefore, a thorough understanding of the parameters for t-SNE is crucial. Interactive exploration is necessary to select parameters and validate results. The data were first concatenated by columns combining all the segments (228) into the big uniform three-dimensional array resulting a shape (68400 x 3) wherein 1140 segments multiplied by segment length 60, which was the actual segment length used in the analysis, resulting 64800 instances. This large amount of data points in t-SNE analysis caused computational challenges and therefore parameter optimization was left for future investigations. In Fig. 9 , t-SNE algorithm was applied with parameter values: perplexity = 30, iterations = 3000, and learning rate = auto. Data accessibility The dataset is made available to the research community to foster further research and development in the field of Human Activity Recognition and related disciplines. It is hosted on a publicly accessible data repository with an open-access data use agreement. This ensures that the dataset complies with the FAIR (Findable, Accessible, Interoperable, and Reusable) principles, enhancing its utility for widespread scientific use. A DOI (Digital Object Identifier) is assigned to the dataset to provide a persistent link to its location and to facilitate citation in academic publications. Researchers wishing to access the dataset can find it at: https://www.kaggle.com/datasets/jarnomatarmaa/sportdata-mts-5/data and are encouraged to cite it as follows: Matarmaa, J. (2023). Sport Activity Dataset - MTS-5. Kaggle. https://doi.org/10.34740/KAGGLE/DS/3512653 . Moreover, metadata describing the dataset, including the data collection methodology, preprocessing steps, and segmentation details, is available alongside the dataset to ensure transparency and reproducibility. 4. Potential Applications Research in Human Activity Recognition (HAR) – The primary application of this dataset is in the field of Human Activity Recognition (HAR), where it can help develop and refine algorithms that recognize and classify diverse types of outdoor sport activities based on physiological and environmental sensor data. Given the dataset's variety in activities and its multivariate nature, it is ideally suited for testing both existing and novel HAR algorithms. Researchers can use this dataset to improve the accuracy of activity recognition systems in uncontrolled environments, a common challenge in the field. The detailed capture of heart rate, speed, and altitude under various real-world conditions provides testing ground for algorithms designed to cope with environmental variability and noise inherent in real-life settings. Sports Performance Analysis – For sports scientists and coaches, this dataset offers a resource for analyzing athletic performance across different sports. By examining variations in heart rate, speed, and altitude data across activities such as walking, running, biking, and skiing, insights can be gained into the physiological demands of each sport. This information can be used to tailor training programs that enhance individual athletes' performance, manage fatigue, and reduce the risk of injury. Moreover, the long-term collection of data allows for the analysis of performance trends over time, which is crucial for planning seasonal training cycles and tapering periods before competitions. Health and Fitness Monitoring – The dataset can also be utilized in health informatics to monitor and improve individual health outcomes. By analyzing data trends, particularly in heart rate and activity levels, it is possible to offer personalized health advice, monitor the effectiveness of fitness programs, and even predict potential health issues before they become severe. This application is particularly important in wearable technology and health apps, where providing users with accurate and personalized feedback can significantly enhance the user's engagement and health outcomes. For example, if heart rate reacts exceptionally to uphill stress it may be considered as an indicator of some health problem. Machine Learning and Data Science Education – Educationally, the dataset could serve as a resource for teaching aspects of machine learning and data science. It provides a practical case study for students to engage with real-world data, applying preprocessing techniques, exploratory data analysis, and machine learning algorithms. The challenges presented by the dataset, such as handling imbalanced data and transforming raw sensor data into meaningful features, offer students valuable firsthand experience that is directly applicable to industry problems. Development of Adaptive Wearable Technologies – Finally, this dataset can aid in the development of adaptive wearable technologies that adjust their behavior based on the user's activity type. For example, smartwatches could use algorithms developed and refined using this dataset to better detect when a user switches from one type of activity to another, optimizing battery life by adjusting GPS and sensor sampling rates accordingly. 5. Study limitations Several study limitations which merit acknowledgment were recognized. Foremost among these is the absence of data from multiple athletes. Given that the dataset originates from a single athlete, the study does not account for interpersonal differences. For instance, the unique behavior of sensor data, such as heart rate, varies among individuals, necessitating separate model training for classification tasks for each person. Therefore, the added value provided by the dataset must not be confused with tasks that seek to generalize sport activity patterns among different athletes. The absence of sport activity datasets indicates the challenges associated with collecting and publishing comprehensive sport activity data. This study is one attempt—albeit incomplete—to contribute to this field and complement the very limited availability of public datasets. In addition of being a one-man study, it should be noted that this dataset primarily focuses on outdoor activities due to technological constraints related to GPS-based speed measurements and air pressure-based altitude measurements. 6. Discussion This study introduced a novel multivariate time series dataset, capturing diverse outdoor sports activities recorded by an individual male athlete. The dataset was introduced in the context of human activity recognition since it was the original motive for collecting the dataset and one of the most relevant application fields, although inertial data has been usually and most successfully used in the context of HAR. It provides detailed, real-world data on heart rate, speed, and altitude recorded under uncontrolled environmental conditions. Its public availability offers a valuable resource for developing and testing systems that require robustness to the variability inherent in natural settings. The dataset's comprehensive scope and the granularity of the collected data allow researchers to explore complex questions related to athletic performance and environmental influences on physical activities. By documenting the preprocessing and structuring processes, this dataset also serves as a model for future data collection efforts in sports science and health informatics. Future research could leverage this dataset to advance the development of personalized training programs and provide an interesting alternative for activity recognition systems in real-world scenarios. Additionally, expanding the dataset to include more participants would increase its generalizability and potential for broader applications. Most likely, this dataset is only the first part of the series, and soon data from numerous athletes representing diverse groups in age, gender, and others will be added. In making this dataset publicly available, this study supports open scientific collaboration and fosters further research into the dynamics of human activity in diverse environmental conditions. We believe that this dataset will be a valuable addition to the research community and will contribute to the advancement of our understanding of outdoor sports and human performance. We encourage researchers and practitioners to explore the dataset and share their findings with the broader community. Declarations Competing interests: The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results. Consent for publication: Informed consent was obtained from all subjects involved in the study. Written informed consent has been obtained from the patient to publish this article. Ethics approval: This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Export Committee of Ural Federal University (protocol 50 − 05 / 1368 from 01.10.2023). Funding: This research was funded by Ministry of Science and Higher Education of the Russian Federation (Ural Federal University Program of Development within the Priority—2030 Program) is gratefully acknowledged Author Contribution All the contributions belongs to the responsible author. Data Availability Availability of data and materials: The data presented in this study are openly available in kaggle.com, at DOI: 10.34740/kaggle/ds/3512653 and in data.world at https://data.world/jamasoftwares/outdoor-sport-activities-mts-5 References Demrozi F, Pravadelli G, Bihorac A, Rashidi P. Human Activity Recognition Using Inertial, Physiological and Environmental Sensors: A Comprehensive Survey. IEEE Access. 2020. 8:210816–210836. https://doi.org/10.1109/ACCESS.2020.3037715 . Lara OD, Labrador MA. A Survey on Human Activity Recognition using Wearable Sensors. IEEE Commun Surv Tutor. 2013. 15:1192–1209. https://doi.org/10.1109/SURV.2012.110112.00192 . Berchtold M, Budde M, Schmidtke HR, Beigl M. An Extensible Modular Recognition Concept That Makes Activity Recognition Practical. In: Dillmann R, Beyerer J, Hanebeck UD, Schultz T (eds) KI 2010: Advances in Artificial Intelligence. Springer Berlin Heidelberg, Berlin, Heidelberg. 2010. pp 400–409. Lara ÓD, Pérez AJ, Labrador MA, Posada JD. Centinela: A human activity recognition system based on acceleration and vital sign data. Pervasive Mob Comput. 2012 8:717–729. https://doi.org/10.1016/j.pmcj.2011.06.004 . Tapia EM, Intille SS, Haskell W, et al., Real-Time Recognition of Physical Activities and Their Intensities Using Wireless Accelerometers and a Heart Rate Monitor. In: 2007 11th IEEE International Symposium on Wearable Computers. IEEE, Boston, MA, USA. 2007. pp 1–4. Foerster F, Smeja M, Fahrenberg J. Detection of posture and motion by accelerometry: a validation study in ambulatory monitoring. Comput Hum Behav. 1999. 15:571–583. https://doi.org/10.1016/S0747-5632(99)00037-0 . Altun K, Barshan B. Daily and Sports Activities Data Set. 2019. https://doi.org/10.21227/AT1V-6F84 . Logacjov A, Ustad A. HAR70+. 2023. https://doi.org/10.24432/C5CW3D . Micucci D, Mobilio M, Napoletano P. UniMiB SHAR: A Dataset for Human Activity Recognition Using Acceleration Data from Smartphones. Appl Sci. 2017. 7:1101. https://doi.org/10.3390/app7101101 . Anguita D, Ghio A, Oneto L, et al. A Public Domain Dataset for Human Activity Recognition using Smartphones. In: ESANN 2013 proceedings. The European Symposium on Artificial Neural Networks, Bruges (Belgium). 2013. pp 437–442. Reyes-Ortiz J, Anguita D, Ghio A, et al., Human Activity Recognition Using Smartphones. 2013. https://doi.org/10.24432/C54S4K . Davis K, Owusu E. Smartphone Dataset for Human Activity Recognition (HAR) in Ambient Assisted Living (AAL). 2016. https://doi.org/10.24432/C5P597 . Clements J. Basic Motions. Accessed 26 May 2024. (2018) Art. 20 GDPR - Right to data portability. In: GDPR.eu. https://gdpr.eu/article-20-right-to-data-portability/ . Accessed 26 May 2024. van der Maaten L, Hinton G. Visualizing Data using t-SNE. Journal of Machine Learning Research. 2008. 2579–2605. Wattenberg M, Viégas F, Johnson I. How to Use t-SNE Effectively. Distill 1: 10.23915/distill.00002 . 2016. https://doi.org/10.23915/distill.00002. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 2 posted Reviews received at journal 19 Dec, 2024 Reviewers agreed at journal 16 Dec, 2024 Reviews received at journal 12 Dec, 2024 Reviewers agreed at journal 12 Dec, 2024 Reviewers invited by journal 11 Dec, 2024 Editor assigned by journal 11 Dec, 2024 Submission checks completed at journal 11 Dec, 2024 First submitted to journal 10 Dec, 2024 You are reading this latest preprint version Show more versions 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-4593851","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[{"code":1,"date":"2024-07-18 20:02:57","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":400544370,"identity":"c3202d0b-1fb1-4a83-8456-364cdb21fcdc","order_by":0,"name":"Jarno Olavi Matarmaa","email":"data:image/png;base64,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","orcid":"","institution":"Ural Federal University","correspondingAuthor":true,"prefix":"","firstName":"Jarno","middleName":"Olavi","lastName":"Matarmaa","suffix":""}],"badges":[],"createdAt":"2024-06-17 11:27:34","currentVersionCode":2,"declarations":"","doi":"10.21203/rs.3.rs-4593851/v2","doiUrl":"https://doi.org/10.21203/rs.3.rs-4593851/v2","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":73526395,"identity":"b5877b31-497e-4ab7-a4d9-2ffaa2d31861","added_by":"auto","created_at":"2025-01-10 21:28:45","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":197355,"visible":true,"origin":"","legend":"\u003cp\u003eData recording devices: a) Garmin FR-920XT sport watch, b) Garmin HRM-Run sensor, and c) HRM-Run sensor mounting demonstration. A chest strap features a module and contact patches that can detect and measure pulse via electric signals emitted by heart. This information is then transmitted to a connected device that records the data during an activity.\u003c/p\u003e","description":"","filename":"Screenshot20250110at10.05.45AM.png","url":"https://assets-eu.researchsquare.com/files/rs-4593851/v2/a2b6a1827f8795de5e96486d.png"},{"id":73526022,"identity":"1bdee7b3-0285-4b27-852d-af2745d0e52f","added_by":"auto","created_at":"2025-01-10 21:20:46","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":68984,"visible":true,"origin":"","legend":"\u003cp\u003eCategory distribution of the sport activities (a), and activity distribution among devices (b). Instances here means segments in augmented data.\u003c/p\u003e","description":"","filename":"Screenshot20250110at10.06.54AM.png","url":"https://assets-eu.researchsquare.com/files/rs-4593851/v2/117107cd9c25c9c8a0130bb3.png"},{"id":73526018,"identity":"bbb1505e-f92b-4f77-9178-c6f82e9aa30e","added_by":"auto","created_at":"2025-01-10 21:20:45","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":76289,"visible":true,"origin":"","legend":"\u003cp\u003eData samples for two random 3-dimensional segments. hr=heart rate, spd=speed, alt=altitude.\u003c/p\u003e","description":"","filename":"Screenshot20250110at10.07.13AM.png","url":"https://assets-eu.researchsquare.com/files/rs-4593851/v2/5f7faf217cdfdad52a6ef4d5.png"},{"id":73526734,"identity":"051eb21e-8c7b-4d7f-95b9-2beca040c707","added_by":"auto","created_at":"2025-01-10 21:36:46","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":80357,"visible":true,"origin":"","legend":"\u003cp\u003eThe violin plot of the speed value distribution by categories.\u003c/p\u003e","description":"","filename":"Screenshot20250110at10.07.31AM.png","url":"https://assets-eu.researchsquare.com/files/rs-4593851/v2/dd028d172bbe56de8559a380.png"},{"id":73525939,"identity":"a47f8ede-c97a-421d-8bed-a42af38561cc","added_by":"auto","created_at":"2025-01-10 21:12:46","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":67614,"visible":true,"origin":"","legend":"\u003cp\u003eThe violin plot of the speed value distribution by categories.\u003c/p\u003e","description":"","filename":"Screenshot20250110at10.07.43AM.png","url":"https://assets-eu.researchsquare.com/files/rs-4593851/v2/3731072a7f2274d6b8754a76.png"},{"id":73525945,"identity":"62716535-5558-4200-a6f7-0dcc42d8c5b2","added_by":"auto","created_at":"2025-01-10 21:12:46","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":73874,"visible":true,"origin":"","legend":"\u003cp\u003eThe violin plot of the altitude value distribution by categories.\u003c/p\u003e","description":"","filename":"Screenshot20250110at10.07.53AM.png","url":"https://assets-eu.researchsquare.com/files/rs-4593851/v2/25d5a72f4d5fa42f240f71b7.png"},{"id":73526024,"identity":"237b52e8-4312-4680-b6fc-04b1d461f62c","added_by":"auto","created_at":"2025-01-10 21:20:46","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":81030,"visible":true,"origin":"","legend":"\u003cp\u003eFeature mean signals by categories\u003c/p\u003e","description":"","filename":"Screenshot20250110at10.08.11AM.png","url":"https://assets-eu.researchsquare.com/files/rs-4593851/v2/6cf837c02a7b4605128626dd.png"},{"id":73525944,"identity":"2e13cdd2-7ebe-4f46-b0e4-ce43f9bb5f81","added_by":"auto","created_at":"2025-01-10 21:12:46","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":100902,"visible":true,"origin":"","legend":"\u003cp\u003eKernel Density Estimation (KDE) plots by category for heart rate and speed features. Skiing and R-Skiing have quite similar kernel density whereas others are more distinctive.\u003c/p\u003e","description":"","filename":"Screenshot20250110at10.08.22AM.png","url":"https://assets-eu.researchsquare.com/files/rs-4593851/v2/8bf18dd3e6ec7ea03e169ab0.png"},{"id":73526029,"identity":"3e9e82ad-0ab2-4ba7-ab9e-f93a974b1da7","added_by":"auto","created_at":"2025-01-10 21:20:46","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":309000,"visible":true,"origin":"","legend":"\u003cp\u003eT-SNE component visualizations of the dataset.\u003c/p\u003e","description":"","filename":"Screenshot20250110at10.08.46AM.png","url":"https://assets-eu.researchsquare.com/files/rs-4593851/v2/ca2fb502850588a455ce7a77.png"},{"id":73526953,"identity":"6807e7bb-3bac-4efe-89ba-c5a9a2995b00","added_by":"auto","created_at":"2025-01-10 21:44:47","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1484270,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4593851/v2/2ae9574c-52ab-4d81-9ac2-fa6ca20105f2.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Novel Multivariate Time Series Dataset of Outdoor Sport Activities","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe research domain of the newly introduced novel sport activity dataset is originally motivated and related, although not limited to the field of Human Activity Recognition (HAR). A field that has witnessed extensive exploration over the last two decades, propelled by the surging popularity and advancement of activity bracelets, smartwatches, and smartphones equipped with inertial sensors for data collection [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. The evolution of deep learning and machine learning algorithms has enabled the integration of HAR into various domains, spanning sports, health, and well-being applications. HAR's overarching goal is to recognize human activities in both controlled and uncontrolled settings.\u003c/p\u003e \u003cp\u003eLara and Labrador (2013) contributed a comprehensive overview to the field of HAR studies, specifically addressing personalized sport activity recognition [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Despite subsequent algorithmic developments, their study remains influential, notably in highlighting the ongoing debate on activity recognition model design. Some studies argue for the construction of specific recognition models for individual differences in age, gender, and weight, necessitating system retraining for each new user [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Conversely, other studies emphasize the need for a monolithic recognition model adaptable across diverse users [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThis debate has given rise to two types of analyses for evaluating activity recognition systems: subject-dependent and subject-independent evaluations [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. In subject-dependent evaluations, a classifier is trained and evaluated for each individual, and the average accuracy across all subjects is computed. In subject-independent evaluations, a single classifier is constructed for all individuals using cross-validation or leave-one-individual-out analysis. Lara and Labrador highlighted practical challenges in retraining systems for each new user, particularly in cases with numerous activities, undesirable activities, or uncooperative subjects. Recognizing that individuals, especially across age groups, may exhibit different activity patterns, they proposed addressing the dichotomy of monolithic and personalized models by creating groups of users with similar characteristics [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. However, the practical implementation of this solution is often constrained by the availability of suitable datasets in HAR studies.\u003c/p\u003e \u003cp\u003eDespite achieving impressive accuracies of up to 99% in controlled experiments, concerns persist regarding the realism and environmental variability of data collection settings. Notably, significant accuracy drops have been observed since the beginning of HAR studies in late 1990s when transitioning from controlled laboratory experiments to uncontrolled, non-laboratory natural environments [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Consequently, there is a growing emphasis on conducting studies in more challenging datasets to ensure comprehensive consideration of all environmental variables.\u003c/p\u003e \u003cp\u003eFor HAR studies, wherein inertial sensor data of three-dimensional-accelerometers are usually applied there are quite numerous notable datasets [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan additionalcitationids=\"CR8 CR9\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], some of them easily accessible as they are indexed by search engines. Some of the most common datasets are listed in Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. However, for example, sport activity recognition does not need to be limited to inertial sensors. Rather, the hypothesis could be set in the following way: can we recognize sports using only physiological and environmental data types, instead of inertial? For instance, what is the response of heart rate when there are environmental changes like derivative of altitude, or even base level of altitude? Not only between different sports, but also the effect of altitude among same category of activities. What about the relationship of heart rate and speed: how much heart rate will increase if the speed increases by 10%? For different type of sports, the response of the heart rate for the changes in speed might significantly differ. This type of dataset could allow us to conduct remarkably another level of studies, although it is limited to outdoor activities due to technological limitations of GPS based speed and air pressure-based altitude measurements.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePublic multivariate time series datasets for Human and Sport Activity Recognition.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDataset\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYear, Study\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eData Type / Tasks\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eClasses\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eInstances\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSensors\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUCI HAR 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2012, [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e],[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClassification, Clustering\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e10299\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2 x 3-axial\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUCI HAR 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2016, [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClassification, Clustering\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5744\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2 x 3-axial\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHAR70+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2023, [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClassification\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2\u0026nbsp;259 597\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2 x 3-axial\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDaily and Sport Activities\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2010, [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClassification, Clustering\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9120\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e9 x 3-axial\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBasicMotions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2000s, [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClassification (Education)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2 x 3-axial\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUniMiB SHAR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2016, [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClassification\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e11 771\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1 x 3-axial\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTo address these critical considerations, this study introduces and publishes a new and unique dataset recorded in a genuine end user environment, and permission for research use has been granted later. Multivariate time series data, which includes multiple variables measured over time, can provide valuable insights into the physiological and environmental factors that influence athletic performance. However, publicly available datasets that capture the complexity of outdoor sports in uncontrolled or controlled environments are either nonexistent or limited. A novel dataset was created that includes 228 outdoor sport activities across five categories: walking, running, skiing, roller-skiing, and biking. The dataset was collected by a non-competitive male athlete using Garmin Forerunner 920XT and Vivosport devices, which recorded heart rate, speed, and altitude data. These devices were selected for their compatibility within the same sport activity tracking ecosystem and their use of different technologies for heart rate measurement, with the Vivosport employing an optical sensor. This choice helps mitigate device-specific bias in data collection. The dataset offers a unique opportunity for researchers and practitioners to explore the relationships between physiological and environmental factors in outdoor sports. The objective is to present a thorough overview of the dataset\u0026rsquo;s structure, the preprocessing and segmentation processes employed, and the substantial potential it holds for multifaceted research in activity recognition, endurance training, and performance analytics. Detailed representation of the dataset and its five distinct outdoor sports activities are provided. Each activity is recorded under natural, uncontrolled conditions, offering a rare glimpse into the dynamics of sports performance outside laboratory settings.\u003c/p\u003e \u003cp\u003eThis dataset is uniquely positioned to facilitate nuanced research into how environmental factors influence sports performance and can help in the development of adaptive training programs tailored to individual physiological responses. However, it is important to recognize that environmental factors associated with certain sports, such as skiing, are significantly influenced by individual differences. Skiing can be performed under various conditions and environments, depending on the athlete's geolocation and personal preferences. This variability does not diminish the value of a one-man study; however, it should be considered when applying the dataset to different contexts or applications. Furthermore, the longitudinal nature of the data collection provides a rich canvas to explore seasonal variations and trends over time, making it an invaluable resource for long-term studies in sports and health disciplines. Moreover, by making this dataset publicly available, it encourages open scientific inquiry and collaborative research efforts that can lead to breakthroughs in predictive modelling and real-time monitoring in sports and health sectors.\u003c/p\u003e"},{"header":"2. Methodology","content":"\u003cp\u003e \u003cem\u003eData collection\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe data collection of activities was recorded over a period of 16 months by a non-competitive male athlete, employing wearable devices equipped with multiple sensors. Each activity was recorded under natural, uncontrolled environmental conditions to capture realistic performance data. The athlete\u0026rsquo;s consistent participation over this extended period allowed for the collection of a comprehensive set of data across different seasons and varying weather conditions and environments. The data recording devices of the dataset are Garmin Forerunner 920XT with HRM-Run sensor as shown in the figure series Fig.\u0026nbsp;1, and Garmin Vivosport activity tracker which has optical wrist-based heart rate measurement technology. Vivosport was used to record a quantitative minority of sports, the majority of them walking activities.\u003c/p\u003e \u003cp\u003eThe study followed a generic data acquisition architecture for HAR system, as depicted in the study [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. This architecture included wearable sensors, as shown in Fig.\u0026nbsp;1, integration devices such as laptops or smartphones, and communication protocols like TCP/IP to transfer data to a local computer or remote cloud storage. Recording the activities using a sport watch is a straightforward process. When the user starts recording the activity, a proper predetermined sport profile is manually selected from the device. Sport profile selection sets the label for sport activity, and therefore this action is prone to misclassification due to human nature. When completing the activity, the user manually stops the recording and saves it to the device\u0026rsquo;s memory. The next step after saving activity to the memory of sport watch is to send it to a smartphone application, the process called synchronization of devices. Another option is to send activity via Wi-Fi connection to the cloud service which was available in the Garmin FR-920XT used in data recording. The challenging part of the data collection appears while accessing the data from the cloud service, in this case \u003cem\u003econnect.garmin.com\u003c/em\u003e. These commercial service providers follows the data handling policies set by governments such as policy of data portability in GDPR [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e], providing full rights of the data to the data owner, but practically they often offer very limited tools or ways for data export. Therefore, a massive data processing was required including conversions from a complex TCX (Training Center XML) file format to a CSV (Comma Separated Value) format. The specific data conversion algorithm for this purpose was developed, after which data pre-processing using traditional Python data frames was allowed.\u003c/p\u003e \u003cp\u003e \u003cem\u003eData dimensions\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe sport tracking devices enabled a maximum sampling rate with a one-second time interval. The dataset encapsulates three primary dimensions measured using distinct sensors embedded in the wearable devices: heart rate, geolocation, and barometric pressure. These sensors provided continuous measurement streams that were then transformed into interpretable metrics:\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eHeart Rate\u003c/strong\u003e \u003cp\u003eCaptured directly from a heart rate sensor, reflecting the athlete's physiological response to the activity. The original values are given in beats per minute (bpm), with an average range typically observed from 40 to 220 bpm, depending on the intensity of the activity.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eSpeed\u003c/strong\u003e \u003cp\u003eDerived from the geolocation data, calculated to show the pace at which the athlete travelled over ground. Measured in meters per second (m/s), with values ranging from 0 (when stationary) up to 20 m/s, particularly noticeable in faster-paced activities like biking.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eAltitude\u003c/strong\u003e \u003cp\u003eObtained from the barometric pressure sensor, adjusted to reflect elevation changes during the activities. Altitude values are recorded in meters (m), reflecting the elevation gains or losses during activities, which could help understanding of terrain impacts on performance.\u003c/p\u003e \u003c/p\u003e \u003cp\u003eThese dimensions were specifically chosen for their relevance to assessing athletic performance and their ability to provide insights into the physical demands of different outdoor sport activities. Heart rate could represent an engine that will provide some speed, whereas their correlation is highly affected by the value and changes in altitude. The stress on the uphill is considerable higher than on the downhill leading to an increased heart rate, if speed remains the same, according to the laws of physics and energy. And the heart rate increases as the speed increases on flat ground. However, in biking activity heart rate usually decreases while speed is increasing on the downhill. But it is also possible, that person amplifies acceleration on the downhill using physical effort, leading to an aggressive speed increasement. This forms a rather fascinating sensor data combination.\u003c/p\u003e \u003cp\u003e \u003cem\u003ePreprocessing and cleaning\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe raw data collected underwent meticulous preprocessing, cleaning, and filtering to ensure high quality and usability.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eData Cleaning\u003c/strong\u003e \u003cp\u003eActivities with missing sensor data or recordings of insufficient length were excluded from the dataset. This step was crucial to maintaining the integrity of the dataset, particularly given its relatively small size.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eSegmentation\u003c/strong\u003e \u003cp\u003eThe data was segmented into 60 seconds intervals. This segmentation process began at the 100-second mark of each recorded activity to avoid initial anomalies and stabilize sensor readings. The choice of starting at 100 seconds and using 60 seconds segment length was based on a combination of visual data analysis and domain-specific knowledge, optimizing for both computational efficiency and classification accuracy of preliminary tests conducted in the dataset.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eStandardization\u003c/strong\u003e \u003cp\u003eAll data was standardized to ensure uniformity across different measures, facilitating more accurate analysis and comparison. Features were standardized by centering them around the mean and scaling to unit variance. The standard score of a sample x was calculated as\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv id=\"Equa\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:z=\\frac{x-\\mu\\:}{\\sigma\\:}$$\u003c/div\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003ewhere \u003cem\u003e\u0026micro;\u003c/em\u003e is the mean of the signal, and \u003cem\u003eσ\u003c/em\u003e is the standard deviation of the signal attribute.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eData Augmentation\u003c/strong\u003e \u003cp\u003eTo address the dataset's limited size and enhance the robustness of subsequent analyses, data augmentation was performed by selecting five consecutive one-minute segments from the same activity. This approach effectively increased the volume of data for each activity type without compromising the natural variability inherent in the athlete's performance.\u003c/p\u003e \u003c/p\u003e \u003cp\u003eThe applied data cleansing function processed time series data for activity classification by first loading specified entries, CSV files into data frames, then configuring their indices as datetime objects to facilitate time-based analysis. Relevant features, heart rate, speed, and altitude, were filtered from each data frame. Data cleaning involves replacing placeholder strings with numerical missing values (NaN), converting all data to a numeric format, and addressing missing values through a series of methods: linear interpolation is used first to fill gaps based on surrounding data; remaining gaps are filled by carrying forward the last known value; and any persistently missing values are set to zero. This thorough cleaning ensures the data's continuity and usability for classification algorithms. The following cleaning steps were applied:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e \u003cem\u003eReplacing placeholders\u003c/em\u003e: Converting 'None' strings to NaN to standardize missing value representation.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cem\u003eEnsuring numeric data\u003c/em\u003e: Transforming all feature values to numeric types, handling any conversion errors gracefully.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cem\u003eInterpolating missing values\u003c/em\u003e: Filling NaN values by linear interpolation using forward-backward filling method to maintain data continuity.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cem\u003eZero filling\u003c/em\u003e: Setting any remaining missing values to zero, ensuring no gaps in the data sequence.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eThe results of pre-processing and filtering are depicted in the Table \u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The final pre-processed dataset comprises 1140 segments, each 60 seconds in length, across the three dimensions of heart rate, speed, and altitude. Each dimension is stored in separate CSV files to maintain clarity and ease of access for analysis purposes. The segmentation of the original processed sequences was implemented using that clean and filtered data these segments are the end product of this study, the public dataset of sport activities in five categories.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSequence/Signal length statistics before and after applying data filtering.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFeature\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOriginal data\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClean data\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eChange %\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ecount\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e280.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e228.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-18.57\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003emean length (s)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4041.621\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4584.307\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;11.86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003estandard deviation\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3077.461\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2793.751\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-10.16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003emin length (s)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e524.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;524.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e25%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1485.750\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2785.500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;87.48\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e50%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3849.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4437.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;15.28\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e75%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5647.250\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5910.500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;4.66\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003emax length (s)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e19781.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e19781.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"3. Results","content":"\u003cp\u003e \u003cem\u003eData structure and format\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe data is stored in four separate CSV files for ease of use and compatibility with most data analysis tools. Each CSV file is named according to the feature type and dimensions, except meta data with label names and device info. For instance, files are named in the format: FEATURE-DATA_std_NxM.csv (e.g., HR-DATA_std_1140x69.csv). So called metadata with sport category labels and recording device names were stored in the fourth file in corresponding indexes. The order of instances in these four files must not be confused when processing them.\u003c/p\u003e \u003cp\u003eThe dataset is structured as a collection of univariate time series, each corresponding to a different dimension of original multivariate dataset. In other words, each time series segment of 1 minute is structured into three files representing the three recorded dimensions: heart rate, speed, and altitude. Each CSV file contains time series data in a transposed matrix format, where columns represent sequential time points and rows correspond to the different sport activities. This transposed format allows us to use traditional data structure for classification tasks which is generally used as a valid input format in the machine learning library algorithms (e.g. Scikit-Learn). Thus, corresponding indexes in each feature file represents the single activity.\u003c/p\u003e \u003cp\u003eThere are several ways for structuring multivariate three-dimensional data in two-dimensional space, but the described one was used as most convenient structure as the dataset is designed to be used specifically in classification tasks, among others.\u003c/p\u003e \u003cp\u003e \u003cem\u003eData characteristics and features\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe dataset consists of 1140 segments, with each segment in length of 60 seconds. The dataset encompasses a total of 228 recorded activities, divided into categories and recording devices according to the Fig.\u0026nbsp;2.\u003c/p\u003e \u003cp\u003eData samples can have many types of patterns and value levels and their combinations. For instance, as we can see from the Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e3\u003c/span\u003e, some random three-dimensional segment A (Activity A) has lower-level speed value and smaller variance, whereas in segment B (Activity B) speed value is fluctuating noticeably more in sixty seconds time window. This could be potential indicator of diverse types of sport, especially when considering heart rate value at the same time window. Altitude with remarkably similar decreasing trend probably indicates that activities are performed in the same environment. That could be one of the ways to interpret and investigate the data.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWe use violin plots to observe the most obvious value range characteristics of the sports as seen in Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e4\u003c/span\u003e, \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e5\u003c/span\u003e, and \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e6\u003c/span\u003e. Violin plots depict effectively how the values are distributed providing more information compared to traditional boxplot. We can detect different kinds of shapes for each category and thus suppose sport specific behaviour for heart rate, speed, and altitude features.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo further illustrate the distinctiveness and characteristics of the data between features and categories we could use very descriptive mean value segments as in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e7\u003c/span\u003e. It is likely the easiest way of understanding the nature of the segmented data. To mention a few, for instance, biking is clearly the fastest sport among all the others and walking the slowest. Also, biking segments has very low hear rate value emphasizing the effectiveness of the sport, as a higher heart rate during the sport activity leads often more rapidly to physical fatigue. Noticeable observation is that altitude feature has smaller gaps among sports which should be considered when conducting any machine learning task in the data.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eKernel density estimations\u003c/em\u003e \u003c/p\u003e \u003cp\u003eGiven that the dataset is three-dimensional, with altitude lacking a clear relationship with speed and heart rate, the latter two attributes are chosen for visualization due to their notable positive correlation. Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e8\u003c/span\u003e illustrates the distribution of speed-heart rate value points across categories. Notably, Skiing and R-Skiing exhibit similar patterns, while other categories demonstrate more discriminative characteristics. When kernel density graphics of categories are amalgamated into a single figure, intersection areas will become more apparent. Density values are normalized by category to account for an uneven distribution of instances.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003et-SNE\u003c/em\u003e \u003c/p\u003e \u003cp\u003eLastly, we applied t-distributed stochastic neighbor embedding algorithm for clustering the data. T-SNE is a statistical method for visualizing multi-dimensional data by giving each datapoint a location in a two or three-dimensional map. [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] While t-SNE plots often appear to display clusters, the visual clusters can be significantly influenced by the chosen parameterization. Therefore, a thorough understanding of the parameters for t-SNE is crucial. Interactive exploration is necessary to select parameters and validate results. The data were first concatenated by columns combining all the segments (228) into the big uniform three-dimensional array resulting a shape (68400 x 3) wherein 1140 segments multiplied by segment length 60, which was the actual segment length used in the analysis, resulting 64800 instances. This large amount of data points in t-SNE analysis caused computational challenges and therefore parameter optimization was left for future investigations. In Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e9\u003c/span\u003e, t-SNE algorithm was applied with parameter values: perplexity\u0026thinsp;=\u0026thinsp;30, iterations\u0026thinsp;=\u0026thinsp;3000, and learning rate\u0026thinsp;=\u0026thinsp;auto.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eData accessibility\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe dataset is made available to the research community to foster further research and development in the field of Human Activity Recognition and related disciplines. It is hosted on a publicly accessible data repository with an open-access data use agreement. This ensures that the dataset complies with the FAIR (Findable, Accessible, Interoperable, and Reusable) principles, enhancing its utility for widespread scientific use.\u003c/p\u003e \u003cp\u003eA DOI (Digital Object Identifier) is assigned to the dataset to provide a persistent link to its location and to facilitate citation in academic publications. Researchers wishing to access the dataset can find it at:\u003c/p\u003e \u003cp\u003e \u003cspan class=\"ExternalRef\"\u003e \u003cspan class=\"RefSource\"\u003ehttps://www.kaggle.com/datasets/jarnomatarmaa/sportdata-mts-5/data\u003c/span\u003e \u003cspan address=\"https://www.kaggle.com/datasets/jarnomatarmaa/sportdata-mts-5/data\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e \u003c/span\u003e \u003c/p\u003e \u003cp\u003eand are encouraged to cite it as follows:\u003c/p\u003e \u003cp\u003eMatarmaa, J. (2023). Sport Activity Dataset - MTS-5. Kaggle. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.34740/KAGGLE/DS/3512653\u003c/span\u003e\u003cspan address=\"10.34740/KAGGLE/DS/3512653\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eMoreover, metadata describing the dataset, including the data collection methodology, preprocessing steps, and segmentation details, is available alongside the dataset to ensure transparency and reproducibility.\u003c/p\u003e"},{"header":"4. Potential Applications","content":"\u003cp\u003e \u003cem\u003eResearch in Human Activity Recognition (HAR)\u003c/em\u003e \u0026ndash; The primary application of this dataset is in the field of Human Activity Recognition (HAR), where it can help develop and refine algorithms that recognize and classify diverse types of outdoor sport activities based on physiological and environmental sensor data. Given the dataset's variety in activities and its multivariate nature, it is ideally suited for testing both existing and novel HAR algorithms. Researchers can use this dataset to improve the accuracy of activity recognition systems in uncontrolled environments, a common challenge in the field. The detailed capture of heart rate, speed, and altitude under various real-world conditions provides testing ground for algorithms designed to cope with environmental variability and noise inherent in real-life settings.\u003c/p\u003e \u003cp\u003e \u003cem\u003eSports Performance Analysis\u003c/em\u003e \u0026ndash; For sports scientists and coaches, this dataset offers a resource for analyzing athletic performance across different sports. By examining variations in heart rate, speed, and altitude data across activities such as walking, running, biking, and skiing, insights can be gained into the physiological demands of each sport. This information can be used to tailor training programs that enhance individual athletes' performance, manage fatigue, and reduce the risk of injury. Moreover, the long-term collection of data allows for the analysis of performance trends over time, which is crucial for planning seasonal training cycles and tapering periods before competitions.\u003c/p\u003e \u003cp\u003e \u003cem\u003eHealth and Fitness Monitoring\u003c/em\u003e \u0026ndash; The dataset can also be utilized in health informatics to monitor and improve individual health outcomes. By analyzing data trends, particularly in heart rate and activity levels, it is possible to offer personalized health advice, monitor the effectiveness of fitness programs, and even predict potential health issues before they become severe. This application is particularly important in wearable technology and health apps, where providing users with accurate and personalized feedback can significantly enhance the user's engagement and health outcomes. For example, if heart rate reacts exceptionally to uphill stress it may be considered as an indicator of some health problem.\u003c/p\u003e \u003cp\u003e \u003cem\u003eMachine Learning and Data Science Education\u003c/em\u003e \u0026ndash; Educationally, the dataset could serve as a resource for teaching aspects of machine learning and data science. It provides a practical case study for students to engage with real-world data, applying preprocessing techniques, exploratory data analysis, and machine learning algorithms. The challenges presented by the dataset, such as handling imbalanced data and transforming raw sensor data into meaningful features, offer students valuable firsthand experience that is directly applicable to industry problems.\u003c/p\u003e \u003cp\u003e \u003cem\u003eDevelopment of Adaptive Wearable Technologies\u003c/em\u003e \u0026ndash; Finally, this dataset can aid in the development of adaptive wearable technologies that adjust their behavior based on the user's activity type. For example, smartwatches could use algorithms developed and refined using this dataset to better detect when a user switches from one type of activity to another, optimizing battery life by adjusting GPS and sensor sampling rates accordingly.\u003c/p\u003e"},{"header":"5. Study limitations","content":"\u003cp\u003eSeveral study limitations which merit acknowledgment were recognized. Foremost among these is the absence of data from multiple athletes. Given that the dataset originates from a single athlete, the study does not account for interpersonal differences. For instance, the unique behavior of sensor data, such as heart rate, varies among individuals, necessitating separate model training for classification tasks for each person. Therefore, the added value provided by the dataset must not be confused with tasks that seek to generalize sport activity patterns among different athletes. The absence of sport activity datasets indicates the challenges associated with collecting and publishing comprehensive sport activity data. This study is one attempt\u0026mdash;albeit incomplete\u0026mdash;to contribute to this field and complement the very limited availability of public datasets. In addition of being a one-man study, it should be noted that this dataset primarily focuses on outdoor activities due to technological constraints related to GPS-based speed measurements and air pressure-based altitude measurements.\u003c/p\u003e"},{"header":"6. Discussion","content":"\u003cp\u003eThis study introduced a novel multivariate time series dataset, capturing diverse outdoor sports activities recorded by an individual male athlete. The dataset was introduced in the context of human activity recognition since it was the original motive for collecting the dataset and one of the most relevant application fields, although inertial data has been usually and most successfully used in the context of HAR. It provides detailed, real-world data on heart rate, speed, and altitude recorded under uncontrolled environmental conditions. Its public availability offers a valuable resource for developing and testing systems that require robustness to the variability inherent in natural settings.\u003c/p\u003e \u003cp\u003eThe dataset's comprehensive scope and the granularity of the collected data allow researchers to explore complex questions related to athletic performance and environmental influences on physical activities. By documenting the preprocessing and structuring processes, this dataset also serves as a model for future data collection efforts in sports science and health informatics. Future research could leverage this dataset to advance the development of personalized training programs and provide an interesting alternative for activity recognition systems in real-world scenarios. Additionally, expanding the dataset to include more participants would increase its generalizability and potential for broader applications. Most likely, this dataset is only the first part of the series, and soon data from numerous athletes representing diverse groups in age, gender, and others will be added.\u003c/p\u003e \u003cp\u003eIn making this dataset publicly available, this study supports open scientific collaboration and fosters further research into the dynamics of human activity in diverse environmental conditions. We believe that this dataset will be a valuable addition to the research community and will contribute to the advancement of our understanding of outdoor sports and human performance. We encourage researchers and practitioners to explore the dataset and share their findings with the broader community.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cb\u003eCompeting interests:\u003c/b\u003e \u003cp\u003eThe funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication:\u003c/strong\u003e \u003cp\u003eInformed consent was obtained from all subjects involved in the study. Written informed consent has been obtained from the patient to publish this article.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eEthics approval:\u003c/strong\u003e \u003cp\u003eThis study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Export Committee of Ural Federal University (protocol 50\u0026thinsp;\u0026minus;\u0026thinsp;05 / 1368 from 01.10.2023).\u003c/p\u003e \u003c/p\u003e\u003cb\u003eFunding:\u003c/b\u003e \u003cp\u003eThis research was funded by Ministry of Science and Higher Education of the Russian Federation (Ural Federal University Program of Development within the Priority\u0026mdash;2030 Program) is gratefully acknowledged\u003c/p\u003e\u003cb\u003eAuthor Contribution\u003c/b\u003e\u003cp\u003eAll the contributions belongs to the responsible author.\u003c/p\u003e\u003cb\u003eData Availability\u003c/b\u003e\u003cp\u003eAvailability of data and materials: The data presented in this study are openly available in kaggle.com, at DOI: 10.34740/kaggle/ds/3512653 and in data.world at https://data.world/jamasoftwares/outdoor-sport-activities-mts-5\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eDemrozi F, Pravadelli G, Bihorac A, Rashidi P. Human Activity Recognition Using Inertial, Physiological and Environmental Sensors: A Comprehensive Survey. IEEE Access. 2020. 8:210816\u0026ndash;210836. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/ACCESS.2020.3037715\u003c/span\u003e\u003cspan address=\"10.1109/ACCESS.2020.3037715\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLara OD, Labrador MA. A Survey on Human Activity Recognition using Wearable Sensors. IEEE Commun Surv Tutor. 2013. 15:1192\u0026ndash;1209. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/SURV.2012.110112.00192\u003c/span\u003e\u003cspan address=\"10.1109/SURV.2012.110112.00192\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBerchtold M, Budde M, Schmidtke HR, Beigl M. An Extensible Modular Recognition Concept That Makes Activity Recognition Practical. In: Dillmann R, Beyerer J, Hanebeck UD, Schultz T (eds) KI 2010: Advances in Artificial Intelligence. Springer Berlin Heidelberg, Berlin, Heidelberg. 2010. pp 400\u0026ndash;409.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLara \u0026Oacute;D, P\u0026eacute;rez AJ, Labrador MA, Posada JD. Centinela: A human activity recognition system based on acceleration and vital sign data. Pervasive Mob Comput. 2012 8:717\u0026ndash;729. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.pmcj.2011.06.004\u003c/span\u003e\u003cspan address=\"10.1016/j.pmcj.2011.06.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTapia EM, Intille SS, Haskell W, et al., Real-Time Recognition of Physical Activities and Their Intensities Using Wireless Accelerometers and a Heart Rate Monitor. In: 2007 11th IEEE International Symposium on Wearable Computers. IEEE, Boston, MA, USA. 2007. pp 1\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFoerster F, Smeja M, Fahrenberg J. Detection of posture and motion by accelerometry: a validation study in ambulatory monitoring. Comput Hum Behav. 1999. 15:571\u0026ndash;583. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S0747-5632(99)00037-0\u003c/span\u003e\u003cspan address=\"10.1016/S0747-5632(99)00037-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAltun K, Barshan B. Daily and Sports Activities Data Set. 2019. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.21227/AT1V-6F84\u003c/span\u003e\u003cspan address=\"10.21227/AT1V-6F84\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLogacjov A, Ustad A. HAR70+. 2023. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.24432/C5CW3D\u003c/span\u003e\u003cspan address=\"10.24432/C5CW3D\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMicucci D, Mobilio M, Napoletano P. UniMiB SHAR: A Dataset for Human Activity Recognition Using Acceleration Data from Smartphones. Appl Sci. 2017. 7:1101. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/app7101101\u003c/span\u003e\u003cspan address=\"10.3390/app7101101\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAnguita D, Ghio A, Oneto L, et al. A Public Domain Dataset for Human Activity Recognition using Smartphones. In: ESANN 2013 proceedings. The European Symposium on Artificial Neural Networks, Bruges (Belgium). 2013. pp 437\u0026ndash;442.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReyes-Ortiz J, Anguita D, Ghio A, et al., Human Activity Recognition Using Smartphones. 2013. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.24432/C54S4K\u003c/span\u003e\u003cspan address=\"10.24432/C54S4K\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDavis K, Owusu E. Smartphone Dataset for Human Activity Recognition (HAR) in Ambient Assisted Living (AAL). 2016. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.24432/C5P597\u003c/span\u003e\u003cspan address=\"10.24432/C5P597\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eClements J. Basic Motions. Accessed 26 May 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e(2018) Art. 20 GDPR - Right to data portability. In: GDPR.eu. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://gdpr.eu/article-20-right-to-data-portability/\u003c/span\u003e\u003cspan address=\"https://gdpr.eu/article-20-right-to-data-portability/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 26 May 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan der Maaten L, Hinton G. Visualizing Data using t-SNE. Journal of Machine Learning Research. 2008. 2579\u0026ndash;2605.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWattenberg M, Vi\u0026eacute;gas F, Johnson I. How to Use t-SNE Effectively. Distill 1:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.23915/distill.00002\u003c/span\u003e\u003cspan address=\"10.23915/distill.00002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. 2016. https://doi.org/10.23915/distill.00002.\u003c/span\u003e\u003c/li\u003e\u003c/ol\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":"discover-data","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"did","sideBox":"Learn more about [Discover Data](https://www.springer.com/44248)","snPcode":"44248","submissionUrl":"https://submission.nature.com/new-submission/44248/3","title":"Discover Data","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"multivariate time series, outdoor sport, sport exercises, sport dataset","lastPublishedDoi":"10.21203/rs.3.rs-4593851/v2","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4593851/v2","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study introduces a novel multivariate time series dataset of 228 outdoor sport activities recorded by individual non-competitive athlete in uncontrolled environments. The dataset includes three features: Heart Rate, Speed, and Altitude, and covers five sport categories: walking, running, skiing, roller-skiing, and biking. The data was collected using two types of Garmin sport watches. The original dataset was carefully pre-processed using typical data cleansing methods such as gaps filling, and value format transformations. Furthermore, activity filtering was implemented for missing sensor value data and using domain knowledge of sport categories. Full length sequences, varying from 10 minutes to several hours, were split into equal length segments, approximately 1 minute. To address the small number of instances data was augmented using several consecutive segments from the same activity. However, only a small part of the whole original data was used as a computational cost\u0026ndash;information gain tradeoff. Three-dimensional dataset is divided into three parts, each dimension to its own comma separated value (CSV) file. The dataset aims to provide a unique resource for researchers and practitioners in the field of sports science, human performance analysis, and activity recognition. It aims to complement the very limited or non-existent publicly available sport activity datasets.\u003c/p\u003e","manuscriptTitle":"A Novel Multivariate Time Series Dataset of Outdoor Sport Activities","msid":"","msnumber":"","nonDraftVersions":[{"code":2,"date":"2025-01-10 21:12:41","doi":"10.21203/rs.3.rs-4593851/v2","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2024-12-19T13:53:19+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"17007236607424756570531077963932972745","date":"2024-12-16T17:20:26+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-12-12T08:10:45+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"117047576230143186319340559196154764640","date":"2024-12-12T07:24:24+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-12-11T14:32:42+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-12-11T14:32:30+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-12-11T14:32:09+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Data","date":"2024-12-10T18:20:53+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-data","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"did","sideBox":"Learn more about [Discover Data](https://www.springer.com/44248)","snPcode":"44248","submissionUrl":"https://submission.nature.com/new-submission/44248/3","title":"Discover Data","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}},{"code":"","date":"2024-09-11 18:51:11","doi":"","editorialEvents":[{"type":"decision","content":"Revision requested","date":"2024-12-06T15:33:16+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-11-29T07:42:05+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"117047576230143186319340559196154764640","date":"2024-11-22T06:41:32+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-10-20T11:11:55+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"67441002019231198440718976920808607261","date":"2024-10-02T01:06:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"17007236607424756570531077963932972745","date":"2024-09-30T18:29:13+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-09-25T02:30:54+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-09-25T01:34:33+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-09-23T04:55:48+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Data","date":"2024-09-11T18:49:46+00:00","index":"","fulltext":""},{"type":"notPreprinted","content":""}],"status":"timeline","journal":{"display":true,"email":"[email protected]","identity":"discover-data","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"did","sideBox":"Learn more about [Discover Data](https://www.springer.com/44248)","snPcode":"44248","submissionUrl":"https://submission.nature.com/new-submission/44248/3","title":"Discover Data","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"4a3bef3e-8646-4be8-aee5-9ba2e5b493f4","owner":[],"postedDate":"January 10th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-01-10T15:02:33+00:00","versionOfRecord":[],"versionCreatedAt":"2025-01-10 21:12:41","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v2","identity":"rs-4593851","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4593851","identity":"rs-4593851","version":["v2"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}