HAR70+W: Integrating Thigh-Back and Wrist Sensor Data for Enhanced Elderly Activity Monitoring

HAR70+W: Integrating Thigh-Back and Wrist Sensor Data for Enhanced Elderly Activity Monitoring | 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 Short Report HAR 70+ W: Integrating Thigh-Back and Wrist Sensor Data for Enhanced Elderly Activity Monitoring Abha Tewari This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6748794/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Human Activity Recognition (HAR) systems for elderly mon- itoring often rely on single-sensor datasets, limiting their ability to cap- ture full-body movement patterns critical for detecting functional de- cline. While datasets like HAR70+ (thigh/back sensors) excel in lower- body activities (e.g., walking, sitting) and WEDA-Fall (wrist sensors) captures arm gestures (e.g., clapping, knocking), no prior work inte- grates these complementary sources for a comprehensive analysis. To bridge this gap, we present HAR70 + W, the first unified HAR frame- work combining both datasets through video-assisted label remapping to resolve inconsistencies. Our integrated approach enables comprehen- sive monitoring of both gross motor activities and fine hand movements, establishing a foundation for robust, multi-sensor HAR systems tailored to aging-related mobility assessment. Experimental results demonstrate improved activity recognition accuracy compared to single-dataset ap- proaches, highlighting its potential for early detection of functional de- cline in elderly populations. Health Economics & Outcomes Research Elderly Monitoring Multi-Sensor Fusion Aging Popula- tion Human Activity Recognition (HAR) Transfer Learning HAR70 + Dataset EDA-Fall Dataset Figures Figure 1 Figure 2 Figure 3 1 Introduction Monitoring human activity in older adults is essential for promoting health and safety. It allows caregivers and healthcare professionals to detect potential risks or early signs of functional decline. This work presents a novel multimodal fusion- based activity recognition system built by combining two advanced datasets: HAR70 + and WEDA-Fall [ 15 ]. The HAR70 + dataset contains accelerometer data collected from sensors worn on the back and thigh of individuals aged 70–95. It captures key daily activities such as walking, sitting, shuffling and stair use. However, it does not include hand-related movements, which can be important indicators of subtle changes in physical function. To address this gap, we incorporated the WEDA-Fall dataset, which uses wrist-mounted Fitbit sensors to record hand-focused activities like clapping, hit- ting a table, or opening a door. A major challenge in combining these datasets was the inconsistency in ac- tivity labels. HAR70 + does not classify hand-based actions, while WEDA-Fall groups similar activities (e.g., stair ascent and descent) under a single label. To resolve this, we analyzed reference videos provided with the WEDA-Fall dataset. This allowed us to refine the labels by distinguishing visually between similar ac- tions (e.g., “walking up stairs” vs. “walking down stairs”). The video review also helped integrate hand-based activities into the broader set of daily movements captured in HAR70+, ensuring consistency across both datasets. This integrated approach provides a more complete picture of elderly activity patterns and supports the development of more accurate and responsive HAR systems. The rest of the paper is organised as follows: Section ?? presents a compre- hensive literature review of two state-of-the-art datasets, HAR70 + and WEDA- FALL, analyzing their respective strengths and limitations. Section 3 outlines the system design, detailing its key phases and architectural framework. Section 4 describes the implementation process, including experimental results, perfor- mance evaluation, and challenges encountered during the study. Finally, Section 5 and Section 6 concludes the paper and highlights potential directions for future research. Please note that the first paragraph of a section or subsection is not indented. The first paragraph that follows a table, figure, equation etc. does not need an indent, either. Subsequent paragraphs, however, are indented. 2 Related work Our research integrates two complementary datasets that capture a diverse range of human activities in elderly populations through different sensor modalities. This multi-sensor approach provides a comprehensive view of daily behaviors, enhancing activity recognition capabilities. One of these datasets is the HAR70 + dataset[ 6 ]. It contains time-series data from accelerometers placed on the back and thigh. These sensors mainly capture movements of the lower body and changes in posture. The dataset includes several common activities such as walking, which shows basic mobility,shuffling, which is a slower form of walking often seen in people with mobility issues, climbing stairs (going up), which has a unique movement pattern and going down stairs, which also shows distinct physical motions. It also records standing, which requires balance, sitting, which is useful for understanding rest and activity balance and lying down, which helps in analyzing rest or sleep patterns. While the HAR70 + dataset is very good at capturing large movements that are important for elderly routines, it has some limitations. Since the sensors are placed on the lower body, it is not able to track upper body movements or small, fine motor activities. Recent research on Human Activity Recognition (HAR) for elderly care has explored various sensor technologies and learning approaches to enhance mon- itoring accuracy.Wearable sensors such as accelerometers and gyroscopes used to track motion. They apply both statistical feature extraction techniques (like mean, standard deviation and entropy) and deep learning models (CNN, LSTM, ANN) demonstrating that deep learning significantly boosts recognition perfor- mance through automatic feature extraction [ 6 ]. In contrast, a multi-sensor fusion approach was adopted, combining wearable sensors with ambient sensors such as infrared, pressure, and motion detectors. Their work emphasizes context-aware computing and traditional machine learn- ing methods like Decision Trees, Naïve Bayes, SVM, and Hidden Markov Mod- els (HMM) to improve the robustness of activity detection in real-world set- tings [ 13 ].A hybrid approach, integrating both sensor-based and vision-based HAR within IoT-enabled smart homes was presented in the paper [ 1 ]. They use spatio-temporal feature integration and explore a broad spectrum of mod- els—including Random Forest, KNN, ANN, LSTM, and Deep Belief Networks (DBN)—highlighting the importance of edge computing for real-time monitor- ing. The studies highlight several limitations in HAR systems for elderly care. These include challenges like data imbalance, sensor noise, and difficulty in dis- tinguishing similar activities, which affect model accuracy. They also point out computational complexity, energy constraints in wearables, and latency issues in IoT systems. Additionally, concerns such as high false alarm rates in fall detec- tion and privacy risks in vision-based monitoring underline the need for secure and efficient HAR solutions [ 6 ] [ 13 ] [ 1 ]. The second dataset, WEDA-Fall dataset comprises time-series accelerometer data,offering sensitivity to arm and hand movements not captured by lower-body sensors. The dataset was collected using a Fitbit Sense smartwatch worn on the wrist, capturing real-time motion data at 50Hz (with downsampled versions at 40Hz, 25Hz, 10Hz, and 5Hz). It includes 25 participants—13 young (ages 20–46) and 12 elderly (ages 77–95),young participants performed both falls and ADLs, while elderly participants performed only ADLs. A total of 619 ADL signals and 350 fall signals were recorded in a controlled setting using 9 sensing points from accelerometer, gyroscope, and orientation sensors. ADLs include walking, jogging, walking up and downstairs, sitting and standing up, stumbling while walking, clapping hands, and opening and closing doors.Safety was ensured with video monitoring and the use of mattresses during fall activities[ ? ]. The WEDA-FALL dataset has been used to evaluate various ML and DL models for human activity and fall detection. Traditional machine learning mod- els include SVM, Random Forest, KNN, and Naïve Bayes. Deep learning ap- proaches feature LSTM, CNN, hybrid CNN-LSTM models, and Transformer- based architectures. These models aim to improve recognition accuracy and real-time performance[ 3 ] [ 11 ]. WEDA-FALL uses only a wrist-worn sensor, miss- ing full-body movements.Falls were staged, which might differ from real-world falls.The number of participants is small, limiting generalization.There is a class imbalance, with fewer falls than daily activities. The acceleration and gyroscope data of a 77-year-old and 95-year-old user were plotted while performing the walking activity in the Fig. 1 and Fig. 2 The wrist sensor data from WEDA-Fall adds important extra information to HAR70 + by showing details about small hand movements and upper-body actions. The videos that come with the WEDA-Fall dataset helped to clearly understand and correctly label the activities. This made sure that the data was consistent when combining both datasets and helped in picking out useful features from the sensor signals. While adaptive transfer methods work well with raw sensor data, using both video and sensors together gives more accurate labeling of activities, especially for elderly-specific actions. Recent surveys highlight transfer learning as a key solution for HAR with sparse or heterogeneous sensor data [ 4 ]. Our work extends these approaches by combining datasets through video-assisted label alignment. 3 System Design and Architecture The system architecture is based on a transfer learning approach to improve the model’s ability to recognize a wide range of activities across the combined dataset. We used a 1D Convolutional Neural Network (CNN) to learn patterns over time from sensor data, and followed a two-phase training process. The over- all architecture of the proposed Human Activity Recognition system is illustrated Fig. 3 , highlighting the integration of heterogeneous sensor data and the transfer learning-based CNN pipeline. In the first phase, the model was pre-trained using the HAR70 + dataset, which includes clear and well-labeled data for basic movements such as walking, sitting, and climbing or descending stairs. This helped the model learn the core features of these common movements. After pre-training, the feature extraction layers were kept unchanged by freezing their weights. We then replaced the final classification layer to focus on more specific hand-related activities from the WEDA-Fall dataset. In this second phase, the model was fine-tuned using a lower learning rate. This allowed it to adjust to the smaller and more detailed hand movements, while still keeping the general motion patterns it learned earlier. This transfer learning approach provided two key advantages: it reduced training time and improved accuracy for activities that had fewer training sam- ples. To evaluate the model, we used several performance measures, including ac- curacy curves and confusion matrices. These results helped us understand which activities the model could recognize well and which needed further improvement. 4 Implementation Details The steps we followed to prepare and combine the two datasets are explained below. To handle the differences between the HAR70 + and WEDA-Fall datasets—such as where sensors were placed, how activities were labeled, and the characteris- tics of the participants—we applied a careful preprocessing process. This helped ensure the data could be meaningfully combined and used together. First, we focused on making the activity labels consistent. Using the refer- ence videos from the WEDA-Fall dataset, we closely examined the movements and separated combined activities into more specific ones. For example, we split "walking up and down stairs" into two separate actions: "walking up" and "walk- ing down." We also added upper-body activities like clapping, table hitting, and opening or closing doors to the HAR70 + dataset, even though these were not originally included. Next, we selected participants and activities in a way that balanced both com- patibility and diversity. We used data from all 15 participants in the HAR70 + dataset to capture detailed lower-body movement patterns using sensors placed on the back and thighs. From the WEDA-Fall dataset, we chose only those sub- jects and activities that matched well with HAR70 + activity types. This helped maintain consistency in both physical movements and participant characteristics when merging the datasets. To deal with the problem of uneven class representation—especially for hand- related activities—we used data augmentation techniques. For underrepresented actions like "clapping hands" and "hitting the table with a hand," we created more samples by slightly changing the timing (temporal shifts) and adjusting the signal strength (amplitude modulation) using controlled factors. This helped in- crease the variety and number of samples for these rare classes. We also used careful sampling strategies to ensure that all activity types were fairly repre- sented in the training data, which is important to prevent the model from be- coming biased toward the more frequent activities. Since the two datasets used different sensor types, we applied special process- ing methods. The HAR70 + dataset gave us accelerometer readings from sensors on the back and thighs, while the WEDA-Fall dataset provided data from a wrist-worn Fitbit device. In WEDA-Fall, the accelerometer data came in string format (such as accel_x_list, accel_y_list, and accel_z_list), so we had to con- vert these into numerical arrays before using them. After converting the data, we normalized it to make sure the values from different sensors were on the same scale. Finally, we reshaped the data to fit the input format needed by our 1D CNN model. We used a 1D Convolutional Neural Network (CNN) as our main model and applied a transfer learning approach to improve how well the model generalizes to different types of activities. In the pretraining phase, we first trained the model on common activities from the HAR70 + dataset like walking, sitting, and going up or down stairs. This helped the model learn basic movement patterns effectively. After this, we froze the early layers of the network to keep the learned features related to general body movements. Then, in the fine-tuning phase, we replaced the final classification layer and trained it on more detailed, hand-focused activities from the WEDA-Fall dataset, such as opening/closing doors, clapping hands, and hitting a table. We used a lower learning rate during this step to allow the model to slowly adjust to these new, more specific activities without forgetting what it had already learned. The technical environment and tools used in our study are summarized in Ta- ble 1. This includes the programming language, deep learning framework, data handling and preprocessing libraries, model architecture components, training strategies, visualization tools, evaluation metrics, and hardware specifications. These components were selected to support efficient data processing, model de- velopment, and performance evaluation. 5 Results and Discussion To understand how our dataset changed during preprocessing, we used various visualization methods to track the distribution of activities. We observed three key stages in this process. In the beginning, the HAR70 + dataset had a natu- rally uneven distribution of activities—common actions like walking and sitting appeared much more often than less frequent, transitional movements. After applying data augmentation, we generated more samples for rare hand-related activities such as clapping and hitting a table. Although these classes increased in size, the overall distribution still showed some imbalance. Finally, by apply- ing a strategic sampling method, we balanced the number of samples across all activity classes. This helped create a uniform distribution, which is important for fair and effective model training. The WEDA-Fall dataset similarly underwent distribution modifications to ensure compatibility with the merged dataset architecture and to maintain class balance across the integrated dataset. We closely monitored training and validation results throughout the training process to understand how well the model was learning and generalizing. The validation accuracy reached a steady value of around 0.81, while the training accuracy improved slowly during the fine-tuning phase. This pattern suggests the model might be underfitting slightly, rather than overfitting. When we looked at the loss values, both training and validation losses steadily decreased, which shows that the model was learning effectively. However, since the validation accuracy didn’t improve much alongside the loss, it may mean the model wasn’t adapting well to the new features. This could be due to several factors, such as keeping the early layers frozen, using a learning rate that was too low, or remaining imbalances in the activity classes. The confusion matrix helped us understand how well the model recognized each activity. Among the different classes, the activity "hittablewithhand" showed excellent results, with the model correctly identifying 2,507 out of about 2,528 cases—almost perfect performance. The activity "openingclosingdoor" was rec- ognized fairly well, but there were some noticeable errors, particularly 454 times when it was confused with "clappinghands." The most difficult activity for the model to classify was "clappinghands." It was often misclassified—829 times—as "openingclosingdoor." This suggests that these two hand-related actions share very similar motion patterns, making it harder for the model to tell them apart. We used quantitative metrics to evaluate how well the model performed. Overall, the model reached a validation accuracy of 81.29%, showing strong per- formance across different activities. When looking at specific activities, the model was very accurate in recognizing "hittablewithhand," achieving high scores for precision, recall, and F1-score—all around 0.99—making it the best-recognized activity. For "openingclosingdoor," the model performed reasonably well, with an F1-score of 0.79. It had a high recall of 0.84, meaning it detected most actual instances correctly, but a lower precision of 0.74, meaning it also made some incorrect guesses. The activity "clappinghands" had the lowest recognition per- formance, with an F1-score of only 0.58 and a recall of 0.51, confirming earlier findings from the confusion matrix that it was often mistaken for "openingclos- ingdoor." These findings underscore both the capabilities and limitations of our multi-sensor integration approach, suggesting directions for future refinements in sensor placement, feature engineering, and model architecture design. Future work could adopt transfer learning techniques [ 4 ] to generalize the multimodal fusion framework to unseen populations These differences in performance across activities give us useful insights. The very high accuracy for "hittablewithhand" is likely because it creates a strong and unique motion pattern in the sensor data, making it easy to identify. On the other hand, "clappinghands" and "openingclosingdoor" produced very similar motion signals when recorded from the wrist, which made them difficult for the model to tell apart. This shows a major limitation of using just one sensor on the wrist—it struggles to distinguish between actions that look similar in mo- tion but are different in meaning. Lastly, although our transfer learning method worked well overall, it was less effective for fine hand movements. This suggests that features learned from larger body movements don’t fully capture the details needed to recognize smaller, more subtle hand-based actions. 6 Conclusion and Future Work This study presents an effective method for combining different types of sensor data to improve activity recognition in elderly individuals. By carefully merging the HAR70 + and WEDA-Fall datasets, aligning sensor data, and balancing ac- tivity classes, we built a transfer learning model that achieved 81.29% validation accuracy across varied activities. Our findings highlight both the strengths and current challenges of multi- sensor fusion. While the model performed well overall, it struggled with similar hand movements. Future improvements should focus on better feature design, smarter data augmentation, and model adjustments to better separate overlap- ping motions. Next steps include using more advanced deep learning models, adding other sensors like gyroscopes, and testing real-time systems for elderly care. This work supports the development of privacy-preserving, noninvasive health monitoring tools that help detect early signs of decline and support independent living. Future work could integrate generative explanatory systems like to automatically annotate rare elderly activities in our multimodal framework[ 16 ] Declarations Declaration of AI Use While writing this paper, the authors used generative AI and AI-based tools to improve the language and make the content easier to understand. The authors take full responsibility for the accuracy, honesty, and ethical quality of the research presented in this work. Ethics Approval and Consent to Participate Not Applicable Consent to Publish Not Applicable References Aarthi S, Juliet S (2021) A Comprehensive Study on Human Activity Recognition. 2021 3rd Int. Conf. Signal Process. Commun. (ICSPC). IEEE, Coim- batore, pp 59–63 Beach J (2023) HARTH Dataset. Kaggle Caroppo A, Manni A, Rescio G, Carluccio AM, Siciliano P, Leone A (2024) A Transfer Learning Approach for an Automatic Fall Detection System Based on the Combined Use of a Smartwatch and a Smartphone. In: Fiorini L, Sorrentino A, Siciliano P, Cavallo F (eds) Ambient Assisted Living. ForItAAL 2024. Lecture Notes in Bioengineering. Springer, Cham. https://doi.org/10.1007/978-3-031-77318-1_23 Dhekane SG, Ploetz T (2025) Transfer learning in sensor-based human activity recognition: A survey, ACM Computing Surveys , vol. 57, no. 8, pp. 1–39. https://doi.org/10.1145/3717608 Fula V, Moreno P (2024) Wrist-Based Fall Detection: Towards Gen- eralization across Datasets, Sensors , vol. 24, no. 5, Mar. https://doi.org/10.3390/s24051679 Hayat A, Morgado-Dias F, Bhuyan BP, Tomar R (2022) Human Activity Recogni- tion for Elderly People Using Machine and Deep Learning Approaches. Information 13(6). https://doi.org/10.3390/info13060275 Logacjov A, Bach K, Kongsvold A, Bårdstu HB, Mork PJ (2021) HARTH: A human activity recognition dataset for machine learning. Sensors 21(23):7853. https://doi.org/10.3390/s21237853 Li Y, Zuo Z, Pan J (2023) Sensor-based fall detection using a combination model of a temporal convolutional network and a gated recurrent unit. Future Gener Comput Syst 139:53–63. https://doi.org/10.1016/j.future.2022.09.011 Ministry of Statistics and Programme Implementation DataViz . Government of India. https://mospi.gov.in/dataviz Mohan D, Al-Hamid DZ, Chong PHJ, Sudheera KLK, Gutier- rez J, Chan HCB, Li H (2024) Artificial Intelligence and IoT in El- derly Fall Prevention: A Review. IEEE Sens J 24(4):4181–4198. https://doi.org/10.1109/JSEN.2023.3344605 Marques J, Moreno P (2023) Online Fall Detection Using Wrist Devices. Sensors 23(3). https://doi.org/10.3390/s23031146 Qin X, Chen Y, Wang J, Yu C (2019) Cross-dataset activity recognition via adaptive spatial-temporal transfer learning. *Proc. ACM Interact. Mob. Wearable Ubiqui- tous Technol.* 3(4), 1–24 https://doi.org/10.1145/3369818 Vasudavan H, Balakrishnan S, Murugesan RK (2021) A Study on Activity Recogni- tion Technology for Elderly Care. Int J Curr Res Rev World Health Organization. Falls. WHO, Fact Sheets (2024) 21 February https://www.who.int/news-room/fact-sheets/detail/falls Xaviar S, Yang X, Ardakanian O (2024) Centaur: Robust multimodal fu- sion for human activity recognition. IEEE Sens J 24(11):18578–18591. https://doi.org/10.1109/JSEN.2024.3388893 Yuan S, Navid SP, Jorge O (2025) GeXSe (Generative Explanatory Sen- sor System): A deep generative method for human activity recogni- tion of smart spaces IoT. *IEEE Sens J * (Early Access. https://doi.org/10.1109/JSEN.2025.3541158 Zhuang F, Qi Z, Duan K, Xi D, Zhu Y, Zhu H, Xiong H, He Q (2019) A com- prehensive survey on transfer learning. *arXiv preprint* arXiv:1911.02685 http://arxiv.org/abs/1911.02685 Zeng H, Nguyen H, Yu B, Meng M (2021) Understanding Human Activity Using Deep Learning Networks with Wearable Sensors. IEEE Trans Ind Inf 17(8):5844–5853 Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6748794","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Short Report","associatedPublications":[],"authors":[{"id":461829330,"identity":"a6dd8b3e-7b37-4995-a5a2-5aded5185e00","order_by":0,"name":"Abha Tewari","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABB0lEQVRIiWNgGAWjYDACCR4Yi7Hx4YcKBgY2qDgBLQlgLc3GEmeAWtiI18LAJsHbxgC3Bifgn9178APjD5vE/tmHGyQk59Xl8ck3MH74wWCRh9OSO+eSJRgS0hJnnEtsMCjcdrgY6DBmyR4GiWKc1tzIMQBqOZzbcIaxIUFy24HENqDDpIFmJTbg0CF/I8f4B0PC/9z5QC0HeOfUgbQw/8anxeBGjhnQlgO5G84wNjbwNjCDtLDhtcXwzhkzi4S05PqNZxibmSWOgfyS2GbZY4Bbi9ztHuMbH2zsjOXOsD//+aGmLk+++fDhGz8q6nBqAYMEVDYjULEBPvV4tI+CUTAKRsEoAAMAhi9UEY9uAbUAAAAASUVORK5CYII=","orcid":"","institution":"sarder patel institute of technology,Mumbai,India","correspondingAuthor":true,"prefix":"","firstName":"Abha","middleName":"","lastName":"Tewari","suffix":""}],"badges":[],"createdAt":"2025-05-26 08:39:13","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":true,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":true},"doi":"10.21203/rs.3.rs-6748794/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6748794/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":83544282,"identity":"35b6c980-2dbc-4747-9518-f331ce159092","added_by":"auto","created_at":"2025-05-28 08:50:11","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":124590,"visible":true,"origin":"","legend":"\u003cp\u003eAcceleration signals captured during walking for users aged 77 and 95 in the WEDA-Fall dataset.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6748794/v1/9cb1787eed552f536003bf6f.png"},{"id":83544283,"identity":"c69d60bc-b8c5-4280-86bd-ea81359ee59f","added_by":"auto","created_at":"2025-05-28 08:50:11","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":134082,"visible":true,"origin":"","legend":"\u003cp\u003eOrientation signals captured during walking for users aged 77 and 95 in WEDA- Fall dataset.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6748794/v1/543adebf584a14d1d5716331.png"},{"id":83543959,"identity":"86f1b1be-2802-4ba1-aa47-16cf0017988e","added_by":"auto","created_at":"2025-05-28 08:42:11","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":57657,"visible":true,"origin":"","legend":"\u003cp\u003eBlock diagram of the proposed HAR system architecture using multi-sensor data and transfer learning.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6748794/v1/f565259df3e3ec56ba5ab5b9.png"},{"id":83544297,"identity":"efae5b89-c161-47a3-950e-ae8e58aae05c","added_by":"auto","created_at":"2025-05-28 08:50:15","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":689046,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6748794/v1/406da5a7-c07e-40b9-9f8f-0d23c16d6651.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eHAR\u003csub\u003e70+\u003c/sub\u003eW: Integrating Thigh-Back and Wrist Sensor Data for Enhanced Elderly Activity Monitoring\u003c/p\u003e","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eMonitoring human activity in older adults is essential for promoting health and safety. It allows caregivers and healthcare professionals to detect potential risks or early signs of functional decline. This work presents a novel multimodal fusion- based activity recognition system built by combining two advanced datasets: HAR70\u0026thinsp;+\u0026thinsp;and WEDA-Fall [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe HAR70\u0026thinsp;+\u0026thinsp;dataset contains accelerometer data collected from sensors worn on the back and thigh of individuals aged 70\u0026ndash;95. It captures key daily activities such as walking, sitting, shuffling and stair use. However, it does not include hand-related movements, which can be important indicators of subtle changes in physical function.\u003c/p\u003e \u003cp\u003eTo address this gap, we incorporated the WEDA-Fall dataset, which uses wrist-mounted Fitbit sensors to record hand-focused activities like clapping, hit- ting a table, or opening a door.\u003c/p\u003e \u003cp\u003eA major challenge in combining these datasets was the inconsistency in ac- tivity labels. HAR70\u0026thinsp;+\u0026thinsp;does not classify hand-based actions, while WEDA-Fall groups similar activities (e.g., stair ascent and descent) under a single label. To resolve this, we analyzed reference videos provided with the WEDA-Fall dataset. This allowed us to refine the labels by distinguishing visually between similar ac- tions (e.g., \u0026ldquo;walking up stairs\u0026rdquo; vs. \u0026ldquo;walking down stairs\u0026rdquo;). The video review also helped integrate hand-based activities into the broader set of daily movements captured in HAR70+, ensuring consistency across both datasets.\u003c/p\u003e \u003cp\u003eThis integrated approach provides a more complete picture of elderly activity patterns and supports the development of more accurate and responsive HAR systems.\u003c/p\u003e \u003cp\u003eThe rest of the paper is organised as follows: Section \u003cb\u003e??\u003c/b\u003e presents a compre- hensive literature review of two state-of-the-art datasets, HAR70\u0026thinsp;+\u0026thinsp;and WEDA- FALL, analyzing their respective strengths and limitations. Section 3 outlines the system design, detailing its key phases and architectural framework. Section 4 describes the implementation process, including experimental results, perfor- mance evaluation, and challenges encountered during the study. Finally, Section 5 and Section 6 concludes the paper and highlights potential directions for future research. Please note that the first paragraph of a section or subsection is not indented. The first paragraph that follows a table, figure, equation etc. does not need an indent, either.\u003c/p\u003e \u003cp\u003eSubsequent paragraphs, however, are indented.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"2 Related work","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eOur research integrates two complementary datasets that capture a diverse range of human activities in elderly populations through different sensor modalities. This multi-sensor approach provides a comprehensive view of daily behaviors, enhancing activity recognition capabilities.\u003c/p\u003e \u003cp\u003eOne of these datasets is the HAR70\u0026thinsp;+\u0026thinsp;dataset[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. It contains time-series data from accelerometers placed on the back and thigh. These sensors mainly capture movements of the lower body and changes in posture. The dataset includes several common activities such as walking, which shows basic mobility,shuffling, which is a slower form of walking often seen in people with mobility issues, climbing stairs (going up), which has a unique movement pattern and going down stairs, which also shows distinct physical motions. It also records standing, which requires balance, sitting, which is useful for understanding rest and activity balance and lying down, which helps in analyzing rest or sleep patterns.\u003c/p\u003e \u003cp\u003eWhile the HAR70\u0026thinsp;+\u0026thinsp;dataset is very good at capturing large movements that are important for elderly routines, it has some limitations. Since the sensors are placed on the lower body, it is not able to track upper body movements or small, fine motor activities.\u003c/p\u003e \u003cp\u003eRecent research on Human Activity Recognition (HAR) for elderly care has explored various sensor technologies and learning approaches to enhance mon- itoring accuracy.Wearable sensors such as accelerometers and gyroscopes used to track motion. They apply both statistical feature extraction techniques (like mean, standard deviation and entropy) and deep learning models (CNN, LSTM, ANN) demonstrating that deep learning significantly boosts recognition perfor- mance through automatic feature extraction [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn contrast, a multi-sensor fusion approach was adopted, combining wearable sensors with ambient sensors such as infrared, pressure, and motion detectors. Their work emphasizes context-aware computing and traditional machine learn- ing methods like Decision Trees, Na\u0026iuml;ve Bayes, SVM, and Hidden Markov Mod- els (HMM) to improve the robustness of activity detection in real-world set- tings [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].A hybrid approach, integrating both sensor-based and vision-based HAR within IoT-enabled smart homes was presented in the paper [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. They use spatio-temporal feature integration and explore a broad spectrum of mod- els\u0026mdash;including Random Forest, KNN, ANN, LSTM, and Deep Belief Networks (DBN)\u0026mdash;highlighting the importance of edge computing for real-time monitor- ing.\u003c/p\u003e \u003cp\u003eThe studies highlight several limitations in HAR systems for elderly care. These include challenges like data imbalance, sensor noise, and difficulty in dis- tinguishing similar activities, which affect model accuracy. They also point out computational complexity, energy constraints in wearables, and latency issues in IoT systems. Additionally, concerns such as high false alarm rates in fall detec- tion and privacy risks in vision-based monitoring underline the need for secure and efficient HAR solutions [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e] [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe second dataset, WEDA-Fall dataset comprises time-series accelerometer data,offering sensitivity to arm and hand movements not captured by lower-body sensors. The dataset was collected using a Fitbit Sense smartwatch worn on the wrist, capturing real-time motion data at 50Hz (with downsampled versions at 40Hz, 25Hz, 10Hz, and 5Hz). It includes 25 participants\u0026mdash;13 young (ages 20\u0026ndash;46) and 12 elderly (ages 77\u0026ndash;95),young participants performed both falls and ADLs, while elderly participants performed only ADLs. A total of 619 ADL signals and 350 fall signals were recorded in a controlled setting using 9 sensing points from accelerometer, gyroscope, and orientation sensors. ADLs include walking, jogging, walking up and downstairs, sitting and standing up, stumbling while walking, clapping hands, and opening and closing doors.Safety was ensured with video monitoring and the use of mattresses during fall activities[\u003cb\u003e?\u003c/b\u003e].\u003c/p\u003e \u003cp\u003eThe WEDA-FALL dataset has been used to evaluate various ML and DL models for human activity and fall detection. Traditional machine learning mod- els include SVM, Random Forest, KNN, and Na\u0026iuml;ve Bayes. Deep learning ap- proaches feature LSTM, CNN, hybrid CNN-LSTM models, and Transformer- based architectures. These models aim to improve recognition accuracy and real-time performance[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. WEDA-FALL uses only a wrist-worn sensor, miss- ing full-body movements.Falls were staged, which might differ from real-world\u003c/p\u003e \u003cp\u003efalls.The number of participants is small, limiting generalization.There is a class imbalance, with fewer falls than daily activities. The acceleration and gyroscope data of a 77-year-old and 95-year-old user were plotted while performing the walking activity in the Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e\u003c/p\u003e \u003cp\u003eThe wrist sensor data from WEDA-Fall adds important extra information to HAR70\u0026thinsp;+\u0026thinsp;by showing details about small hand movements and upper-body actions. The videos that come with the WEDA-Fall dataset helped to clearly understand and correctly label the activities. This made sure that the data was consistent when combining both datasets and helped in picking out useful features from the sensor signals.\u003c/p\u003e \u003cp\u003eWhile adaptive transfer methods work well with raw sensor data, using both video and sensors together gives more accurate labeling of activities, especially for elderly-specific actions. Recent surveys highlight transfer learning as a key solution for HAR with sparse or heterogeneous sensor data [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Our work extends these approaches by combining datasets through video-assisted label alignment.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"3 System Design and Architecture","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe system architecture is based on a transfer learning approach to improve the model\u0026rsquo;s ability to recognize a wide range of activities across the combined dataset. We used a 1D Convolutional Neural Network (CNN) to learn patterns over time from sensor data, and followed a two-phase training process. The over- all architecture of the proposed Human Activity Recognition system is illustrated Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, highlighting the integration of heterogeneous sensor data and the transfer learning-based CNN pipeline.\u003c/p\u003e \u003cp\u003eIn the first phase, the model was pre-trained using the HAR70\u0026thinsp;+\u0026thinsp;dataset, which includes clear and well-labeled data for basic movements such as walking, sitting, and climbing or descending stairs. This helped the model learn the core features of these common movements.\u003c/p\u003e \u003cp\u003eAfter pre-training, the feature extraction layers were kept unchanged by freezing their weights. We then replaced the final classification layer to focus on more specific hand-related activities from the WEDA-Fall dataset. In this second phase, the model was fine-tuned using a lower learning rate. This allowed it to adjust to the smaller and more detailed hand movements, while still keeping the general motion patterns it learned earlier.\u003c/p\u003e \u003cp\u003eThis transfer learning approach provided two key advantages: it reduced training time and improved accuracy for activities that had fewer training sam- ples. To evaluate the model, we used several performance measures, including ac- curacy curves and confusion matrices. These results helped us understand which activities the model could recognize well and which needed further improvement.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"4 Implementation Details","content":"\u003cp\u003eThe steps we followed to prepare and combine the two datasets are explained below.\u003c/p\u003e\u003cp\u003eTo handle the differences between the HAR70\u0026thinsp;+\u0026thinsp;and WEDA-Fall datasets\u0026mdash;such as where sensors were placed, how activities were labeled, and the characteris- tics of the participants\u0026mdash;we applied a careful preprocessing process. This helped ensure the data could be meaningfully combined and used together.\u003c/p\u003e \u003cp\u003eFirst, we focused on making the activity labels consistent. Using the refer- ence videos from the WEDA-Fall dataset, we closely examined the movements and separated combined activities into more specific ones. For example, we split \"walking up and down stairs\" into two separate actions: \"walking up\" and \"walk- ing down.\" We also added upper-body activities like clapping, table hitting, and opening or closing doors to the HAR70\u0026thinsp;+\u0026thinsp;dataset, even though these were not originally included.\u003c/p\u003e \u003cp\u003eNext, we selected participants and activities in a way that balanced both com- patibility and diversity. We used data from all 15 participants in the HAR70\u0026thinsp;+\u0026thinsp;dataset to capture detailed lower-body movement patterns using sensors placed on the back and thighs. From the WEDA-Fall dataset, we chose only those sub- jects and activities that matched well with HAR70\u0026thinsp;+\u0026thinsp;activity types. This helped maintain consistency in both physical movements and participant characteristics when merging the datasets.\u003c/p\u003e\u003cp\u003eTo deal with the problem of uneven class representation\u0026mdash;especially for hand- related activities\u0026mdash;we used data augmentation techniques. For underrepresented actions like \"clapping hands\" and \"hitting the table with a hand,\" we created more samples by slightly changing the timing (temporal shifts) and adjusting the signal strength (amplitude modulation) using controlled factors. This helped in- crease the variety and number of samples for these rare classes. We also used careful sampling strategies to ensure that all activity types were fairly repre- sented in the training data, which is important to prevent the model from be- coming biased toward the more frequent activities.\u003c/p\u003e \u003cp\u003eSince the two datasets used different sensor types, we applied special process- ing methods. The HAR70\u0026thinsp;+\u0026thinsp;dataset gave us accelerometer readings from sensors on the back and thighs, while the WEDA-Fall dataset provided data from a wrist-worn Fitbit device. In WEDA-Fall, the accelerometer data came in string format (such as accel_x_list, accel_y_list, and accel_z_list), so we had to con- vert these into numerical arrays before using them. After converting the data, we normalized it to make sure the values from different sensors were on the same scale. Finally, we reshaped the data to fit the input format needed by our 1D CNN model.\u003c/p\u003e \u003cp\u003eWe used a 1D Convolutional Neural Network (CNN) as our main model and applied a transfer learning approach to improve how well the model generalizes to different types of activities.\u003c/p\u003e \u003cp\u003eIn the pretraining phase, we first trained the model on common activities from the HAR70\u0026thinsp;+\u0026thinsp;dataset like walking, sitting, and going up or down stairs. This helped the model learn basic movement patterns effectively. After this, we froze the early layers of the network to keep the learned features related to general body movements. Then, in the fine-tuning phase, we replaced the final classification layer and trained it on more detailed, hand-focused activities from the WEDA-Fall dataset, such as opening/closing doors, clapping hands, and hitting a table. We used a lower learning rate during this step to allow the model to slowly adjust to these new, more specific activities without forgetting what it had already learned.\u003c/p\u003e \u003cp\u003eThe technical environment and tools used in our study are summarized in Ta- ble 1. This includes the programming language, deep learning framework, data handling and preprocessing libraries, model architecture components, training strategies, visualization tools, evaluation metrics, and hardware specifications. These components were selected to support efficient data processing, model de- velopment, and performance evaluation.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" height=\"590\" width=\"584\"\u003e\u003c/p\u003e"},{"header":"5 Results and Discussion","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eTo understand how our dataset changed during preprocessing, we used various visualization methods to track the distribution of activities. We observed three key stages in this process. In the beginning, the HAR70\u0026thinsp;+\u0026thinsp;dataset had a natu- rally uneven distribution of activities\u0026mdash;common actions like walking and sitting\u003c/p\u003e \u003cp\u003eappeared much more often than less frequent, transitional movements. After applying data augmentation, we generated more samples for rare hand-related activities such as clapping and hitting a table. Although these classes increased in size, the overall distribution still showed some imbalance. Finally, by apply- ing a strategic sampling method, we balanced the number of samples across all activity classes. This helped create a uniform distribution, which is important for fair and effective model training.\u003c/p\u003e \u003cp\u003eThe WEDA-Fall dataset similarly underwent distribution modifications to ensure compatibility with the merged dataset architecture and to maintain class balance across the integrated dataset.\u003c/p\u003e \u003cp\u003eWe closely monitored training and validation results throughout the training process to understand how well the model was learning and generalizing. The validation accuracy reached a steady value of around 0.81, while the training accuracy improved slowly during the fine-tuning phase. This pattern suggests the model might be underfitting slightly, rather than overfitting. When we looked at the loss values, both training and validation losses steadily decreased, which shows that the model was learning effectively. However, since the validation accuracy didn\u0026rsquo;t improve much alongside the loss, it may mean the model wasn\u0026rsquo;t adapting well to the new features. This could be due to several factors, such as keeping the early layers frozen, using a learning rate that was too low, or remaining imbalances in the activity classes.\u003c/p\u003e \u003cp\u003eThe confusion matrix helped us understand how well the model recognized each activity. Among the different classes, the activity \"hittablewithhand\" showed excellent results, with the model correctly identifying 2,507 out of about 2,528 cases\u0026mdash;almost perfect performance. The activity \"openingclosingdoor\" was rec- ognized fairly well, but there were some noticeable errors, particularly 454 times when it was confused with \"clappinghands.\" The most difficult activity for the model to classify was \"clappinghands.\" It was often misclassified\u0026mdash;829 times\u0026mdash;as \"openingclosingdoor.\" This suggests that these two hand-related actions share very similar motion patterns, making it harder for the model to tell them apart.\u003c/p\u003e \u003cp\u003eWe used quantitative metrics to evaluate how well the model performed. Overall, the model reached a validation accuracy of 81.29%, showing strong per- formance across different activities. When looking at specific activities, the model was very accurate in recognizing \"hittablewithhand,\" achieving high scores for precision, recall, and F1-score\u0026mdash;all around 0.99\u0026mdash;making it the best-recognized activity. For \"openingclosingdoor,\" the model performed reasonably well, with an F1-score of 0.79. It had a high recall of 0.84, meaning it detected most actual instances correctly, but a lower precision of 0.74, meaning it also made some incorrect guesses. The activity \"clappinghands\" had the lowest recognition per- formance, with an F1-score of only 0.58 and a recall of 0.51, confirming earlier findings from the confusion matrix that it was often mistaken for \"openingclos- ingdoor.\" These findings underscore both the capabilities and limitations of our multi-sensor integration approach, suggesting directions for future refinements in sensor placement, feature engineering, and model architecture design. Future work could adopt transfer learning techniques [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] to generalize the multimodal fusion framework to unseen populations\u003c/p\u003e \u003cp\u003eThese differences in performance across activities give us useful insights. The very high accuracy for \"hittablewithhand\" is likely because it creates a strong and unique motion pattern in the sensor data, making it easy to identify. On the other hand, \"clappinghands\" and \"openingclosingdoor\" produced very similar motion signals when recorded from the wrist, which made them difficult for the model to tell apart. This shows a major limitation of using just one sensor on the wrist\u0026mdash;it struggles to distinguish between actions that look similar in mo- tion but are different in meaning. Lastly, although our transfer learning method worked well overall, it was less effective for fine hand movements. This suggests that features learned from larger body movements don\u0026rsquo;t fully capture the details needed to recognize smaller, more subtle hand-based actions.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"6 Conclusion and Future Work","content":"\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThis study presents an effective method for combining different types of sensor data to improve activity recognition in elderly individuals. By carefully merging the HAR70\u0026thinsp;+\u0026thinsp;and WEDA-Fall datasets, aligning sensor data, and balancing ac- tivity classes, we built a transfer learning model that achieved 81.29% validation accuracy across varied activities.\u003c/p\u003e\u003cp\u003eOur findings highlight both the strengths and current challenges of multi- sensor fusion. While the model performed well overall, it struggled with similar hand movements. Future improvements should focus on better feature design, smarter data augmentation, and model adjustments to better separate overlap- ping motions.\u003c/p\u003e\u003cp\u003eNext steps include using more advanced deep learning models, adding other sensors like gyroscopes, and testing real-time systems for elderly care. This work supports the development of privacy-preserving, noninvasive health monitoring tools that help detect early signs of decline and support independent living. Future work could integrate generative explanatory systems like to automatically annotate rare elderly activities in our multimodal framework[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eDeclaration of AI Use\u0026nbsp;\u003c/strong\u003eWhile writing this paper, the authors used generative AI and AI-based tools to improve the language and make the content easier to understand. The authors take full responsibility for the accuracy, honesty, and ethical quality of the research presented in this work.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics Approval and Consent to Participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot\u0026nbsp;Applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to Publish\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot Applicable\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAarthi S, Juliet S (2021) A Comprehensive Study on Human Activity Recognition. 2021 3rd Int. Conf. Signal Process. Commun. (ICSPC). IEEE, Coim- batore, pp 59\u0026ndash;63\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBeach J (2023) HARTH Dataset. Kaggle\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCaroppo A, Manni A, Rescio G, Carluccio AM, Siciliano P, Leone A (2024) A Transfer Learning Approach for an Automatic Fall Detection System Based on the Combined Use of a Smartwatch and a Smartphone. In: Fiorini L, Sorrentino A, Siciliano P, Cavallo F (eds) Ambient Assisted Living. ForItAAL 2024. Lecture Notes in Bioengineering. Springer, Cham. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-3-031-77318-1_23\u003c/span\u003e\u003cspan address=\"10.1007/978-3-031-77318-1_23\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDhekane SG, Ploetz T (2025) Transfer learning in sensor-based human activity recognition: A survey, \u003cem\u003eACM Computing Surveys\u003c/em\u003e, vol. 57, no. 8, pp. 1\u0026ndash;39. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/3717608\u003c/span\u003e\u003cspan address=\"10.1145/3717608\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFula V, Moreno P (2024) Wrist-Based Fall Detection: Towards Gen- eralization across Datasets, \u003cem\u003eSensors\u003c/em\u003e, vol. 24, no. 5, Mar. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/s24051679\u003c/span\u003e\u003cspan address=\"10.3390/s24051679\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHayat A, Morgado-Dias F, Bhuyan BP, Tomar R (2022) Human Activity Recogni- tion for Elderly People Using Machine and Deep Learning Approaches. Information 13(6). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/info13060275\u003c/span\u003e\u003cspan address=\"10.3390/info13060275\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLogacjov A, Bach K, Kongsvold A, B\u0026aring;rdstu HB, Mork PJ (2021) HARTH: A human activity recognition dataset for machine learning. Sensors 21(23):7853. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/s21237853\u003c/span\u003e\u003cspan address=\"10.3390/s21237853\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi Y, Zuo Z, Pan J (2023) Sensor-based fall detection using a combination model of a temporal convolutional network and a gated recurrent unit. Future Gener Comput Syst 139:53\u0026ndash;63. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.future.2022.09.011\u003c/span\u003e\u003cspan address=\"10.1016/j.future.2022.09.011\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMinistry of Statistics and Programme Implementation \u003cem\u003eDataViz\u003c/em\u003e. Government of India. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://mospi.gov.in/dataviz\u003c/span\u003e\u003cspan address=\"https://mospi.gov.in/dataviz\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMohan D, Al-Hamid DZ, Chong PHJ, Sudheera KLK, Gutier- rez J, Chan HCB, Li H (2024) Artificial Intelligence and IoT in El- derly Fall Prevention: A Review. IEEE Sens J 24(4):4181\u0026ndash;4198. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/JSEN.2023.3344605\u003c/span\u003e\u003cspan address=\"10.1109/JSEN.2023.3344605\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarques J, Moreno P (2023) Online Fall Detection Using Wrist Devices. Sensors 23(3). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/s23031146\u003c/span\u003e\u003cspan address=\"10.3390/s23031146\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQin X, Chen Y, Wang J, Yu C (2019) Cross-dataset activity recognition via adaptive spatial-temporal transfer learning. *Proc. ACM Interact. Mob. Wearable Ubiqui- tous Technol.* 3(4), 1\u0026ndash;24 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/3369818\u003c/span\u003e\u003cspan address=\"10.1145/3369818\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVasudavan H, Balakrishnan S, Murugesan RK (2021) A Study on Activity Recogni- tion Technology for Elderly Care. Int J Curr Res Rev\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWorld Health Organization. Falls. WHO, Fact Sheets (2024) 21 February \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.who.int/news-room/fact-sheets/detail/falls\u003c/span\u003e\u003cspan address=\"https://www.who.int/news-room/fact-sheets/detail/falls\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXaviar S, Yang X, Ardakanian O (2024) Centaur: Robust multimodal fu- sion for human activity recognition. IEEE Sens J 24(11):18578\u0026ndash;18591. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/JSEN.2024.3388893\u003c/span\u003e\u003cspan address=\"10.1109/JSEN.2024.3388893\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYuan S, Navid SP, Jorge O (2025) GeXSe (Generative Explanatory Sen- sor System): A deep generative method for human activity recogni- tion of smart spaces IoT. *IEEE Sens J * (Early Access. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/JSEN.2025.3541158\u003c/span\u003e\u003cspan address=\"10.1109/JSEN.2025.3541158\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhuang F, Qi Z, Duan K, Xi D, Zhu Y, Zhu H, Xiong H, He Q (2019) A com- prehensive survey on transfer learning. *arXiv preprint* arXiv:1911.02685 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://arxiv.org/abs/1911.02685\u003c/span\u003e\u003cspan address=\"http://arxiv.org/abs/1911.02685\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZeng H, Nguyen H, Yu B, Meng M (2021) Understanding Human Activity Using Deep Learning Networks with Wearable Sensors. IEEE Trans Ind Inf 17(8):5844\u0026ndash;5853\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"sarder patel institute of technology,mumbai,India","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"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},"keywords":"Elderly Monitoring, Multi-Sensor Fusion, Aging Popula- tion, Human Activity Recognition (HAR), Transfer Learning, HAR70 + Dataset, EDA-Fall Dataset","lastPublishedDoi":"10.21203/rs.3.rs-6748794/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6748794/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eHuman Activity Recognition (HAR) systems for elderly mon- itoring often rely on single-sensor datasets, limiting their ability to cap- ture full-body movement patterns critical for detecting functional de- cline. While datasets like HAR70+ (thigh/back sensors) excel in lower- body activities (e.g., walking, sitting) and WEDA-Fall (wrist sensors) captures arm gestures (e.g., clapping, knocking), no prior work inte- grates these complementary sources for a comprehensive analysis. To bridge this gap, we present HAR70\u0026thinsp;+\u0026thinsp;W, the first unified HAR frame- work combining both datasets through video-assisted label remapping to resolve inconsistencies. Our integrated approach enables comprehen- sive monitoring of both gross motor activities and fine hand movements, establishing a foundation for robust, multi-sensor HAR systems tailored to aging-related mobility assessment. Experimental results demonstrate improved activity recognition accuracy compared to single-dataset ap- proaches, highlighting its potential for early detection of functional de- cline in elderly populations.\u003c/p\u003e","manuscriptTitle":"HAR70+W: Integrating Thigh-Back and Wrist Sensor Data for Enhanced Elderly Activity Monitoring","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-28 08:42:06","doi":"10.21203/rs.3.rs-6748794/v1","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}}],"origin":"","ownerIdentity":"e3ec5fd9-2dac-425d-bd14-9cd90b75907d","owner":[],"postedDate":"May 28th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":49156443,"name":"Health Economics \u0026 Outcomes Research"}],"tags":[],"updatedAt":"2025-05-28T08:42:06+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-28 08:42:06","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6748794","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6748794","identity":"rs-6748794","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}