Large Language Models for Wearable Sensor-Based Human Activity Recognition, Health Monitoring, and Behavioral Modeling: A Survey of Early Trends, Datasets, and Challenges
Abstract
:1. Introduction
1.1. Background on Wearable Sensors
1.2. Importance of Data Modeling in Wearable Technology
1.3. Emergence of Large Language Models (LLMs) in Data Analysis
2. Wearable Sensor Data
2.1. Types of Wearable Sensors
2.2. Nature of Data Generated
2.3. Common Applications
Application | Description | Refs. |
---|---|---|
Activity Recognition | Human activity recognition (HAR) is one of the most prominent applications of wearable sensors. By analyzing data from motion sensors, researchers can classify various physical activities, which is valuable in fitness tracking, rehabilitation, and elder care. LLM models like LLaSA and HARGPT have enhanced the accuracy and capabilities of HAR systems. | [12,22,23,24,25,26] |
Health Monitoring | Wearable sensors play a crucial role in continuous health monitoring, enabling the early detection of medical conditions and the management of chronic diseases. Systems like PhysioLLM leverage wearable sensor data to provide personalized health insights and interventions. | [3,13,16,27,28,29,30,31] |
Mental Health | Wearable sensors are increasingly used in mental health applications to monitor physiological indicators of stress, anxiety, and depression. Real-time data from these sensors can be used to develop personalized interventions and support mental well-being. Studies like TILES-2018 and TILES-2019 provide comprehensive datasets that support these applications. MindShift [21] demonstrates LLMs’s ability to generate personalized content using sensor-based data from users’ physical contexts and mental states. | [5,6,21] |
Sports and Fitness | In sports science, wearable sensors are used to monitor athletes’ performance, track training progress, and prevent injuries. The integration of physiological and motion sensors provides comprehensive insights into an athlete’s condition and performance. Advanced coaching systems utilizing LLMs, such as those integrating behavior science principles, have shown significant improvements in training effectiveness. | [4,32,33,34] |
Workplace Ergonomics | Wearable sensors are employed to improve workplace ergonomics by monitoring workers’ movements and posture. These data help in designing ergonomic interventions to prevent musculoskeletal disorders and enhance productivity. | [16,35,36,37] |
3. Large Language Models (LLMs) for Wearable Sensor Data
3.1. Overview of Recent LLM-Based Systems
3.2. Capabilities and Limitations
4. Case Studies and Applications
4.1. Human Activity Recognition
4.2. Health Monitoring
4.3. Mental Health
4.4. Sports Science
4.5. Workplace Ergonomics
4.6. Recent Advancements and Integrations with Other AI Techniques
5. Challenges in Using LLMs for Wearable Sensor Data
5.1. General Issues Related to Using LLMs with Wearable Sensor Data
5.1.1. Data Quality and Preprocessing
5.1.2. Computational Requirements
5.1.3. Interpretability and Transparency
5.2. Specific Issues Related to Using LLMs for Sensor-Based HAR
5.2.1. Data Processing and Integration Complexities
5.2.2. Real-Time Adaptability
5.2.3. Bias and Fairness
5.3. Specific Issues Related to Using LLMs for Sensor-Based Health Monitoring and Mental Health
5.3.1. Health Monitoring
5.3.2. Mental Health
5.4. Specific Issues Related to Using LLMs for Sensor-Based Behavioral Modeling for Sports, Fitness, and Ergonomics
5.4.1. Sports and Fitness
5.4.2. Workplace Ergonomics
6. Ethical and Legal Issues
6.1. User Privacy Protection
6.2. Data Security
6.3. Model Bias
7. Future Directions
7.1. Potential Improvements in LLMs
7.2. Emerging Trends in Wearable Technology
7.3. Interdisciplinary Research Opportunities
8. Conclusions
8.1. Summary of the State of the Art
8.2. Challenges, Trends, and Future Directions
Funding
Data Availability Statement
Conflicts of Interest
References
- Ortiz, B.L. Data Preprocessing Techniques for Artificial Learning (AI)/Machine Learning (ML)-Readiness: Systematic Review of Wearable Sensor Data in Cancer Care. JMIR Mhealth Uhealth 2024. [Google Scholar] [CrossRef]
- Kline, A.; Wang, H.; Li, Y.; Dennis, S.; Hutch, M.; Xu, Z.; Wang, F.; Cheng, F.; Luo, Y. Multimodal machine learning in precision health: A scoping review. NPJ Digit. Med. 2022, 5, 171. [Google Scholar] [CrossRef] [PubMed]
- Fang, C.M.; Danry, V.; Whitmore, N.; Bao, A.; Hutchison, A.; Pierce, C.; Maes, P. PhysioLLM: Supporting Personalized Health Insights with Wearables and Large Language Models. arXiv 2024, arXiv:2406.19283. [Google Scholar]
- Imran, S.A.; Khan, M.N.H.; Biswas, S.; Islam, B. LLaSA: Large Multimodal Agent for Human Activity Analysis through Wearable Sensors. arXiv 2024, arXiv:2406.14498. [Google Scholar]
- Mundnich, K.; Booth, B.M.; l’Hommedieu, M.; Feng, T.; Girault, B.; L’hommedieu, J.; Wildman, M.; Skaaden, S.; Nadarajan, A.; Villatte, J.L.; et al. TILES-2018, a longitudinal physiologic and behavioral data set of hospital workers. Sci. Data 2020, 7, 354. [Google Scholar] [CrossRef]
- Yau, J.C.; Girault, B.; Feng, T.; Mundnich, K.; Nadarajan, A.; Booth, B.M.; Ferrara, E.; Lerman, K.; Hsieh, E.; Narayanan, S. TILES-2019: A longitudinal physiologic and behavioral data set of medical residents in an intensive care unit. Sci. Data 2022, 9, 536. [Google Scholar] [CrossRef]
- Burghardt, K.; Tavabi, N.; Ferrara, E.; Narayanan, S.; Lerman, K. Having a bad day? detecting the impact of atypical events using wearable sensors. In Proceedings of the Social, Cultural, and Behavioral Modeling: 14th International Conference, SBP-BRiMS 2021, Virtual Event, 6–9 July 2021; Proceedings 14; Springer International Publishing: Berlin/Heidelberg, Germany, 2021; pp. 257–267. [Google Scholar]
- Kao, H.T.; Yan, S.; Hosseinmardi, H.; Narayanan, S.; Lerman, K.; Ferrara, E. User-based collaborative filtering mobile health system. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 2020, 4, 1–17. [Google Scholar] [CrossRef]
- Ramanujam, E.; Perumal, T.; Padmavathi, S. Human activity recognition with smartphone and wearable sensors using deep learning techniques: A review. IEEE Sens. J. 2021, 21, 13029–13040. [Google Scholar] [CrossRef]
- Tavabi, N.; Hosseinmardi, H.; Villatte, J.L.; Abeliuk, A.; Narayanan, S.; Ferrara, E.; Lerman, K. Learning Behavioral Representations from Wearable Sensors. In Proceedings of the Social, Cultural, and Behavioral Modeling: 13th International Conference, SBP-BRiMS 2020, Washington, DC, USA, 18–21 October 2020; Proceedings; Springer Nature: Cham, Switzerland, 2020; Volume 12268, p. 245. [Google Scholar]
- Zhang, S.; Li, Y.; Zhang, S.; Shahabi, F.; Xia, S.; Deng, Y.; Alshurafa, N. Deep learning in human activity recognition with wearable sensors: A review on advances. Sensors 2022, 22, 1476. [Google Scholar] [CrossRef]
- Ji, S.; Zheng, X.; Wu, C. HARGPT: Are LLMs Zero-Shot Human Activity Recognizers? arXiv 2024, arXiv:2403.02727. [Google Scholar]
- Kim, Y.; Xu, X.; McDuff, D.; Breazeal, C.; Park, H.W. Health-llm: Large language models for health prediction via wearable sensor data. arXiv 2024, arXiv:2401.06866. [Google Scholar]
- Wang, H.; Zhao, J.; Li, J.; Tian, L.; Tu, P.; Cao, T.; An, Y.; Wang, K.; Li, S. Wearable Sensor-Based Human Activity Recognition Using Hybrid Deep Learning Techniques. Secur. Commun. Netw. 2020, 2020, 2132138. [Google Scholar] [CrossRef]
- McQuire, J.; Watson, P.; Wright, N.; Hiden, H.; Catt, M. A Data Efficient Vision Transformer for Robust Human Activity Recognition from the Spectrograms of Wearable Sensor Data. In Proceedings of the 2023 IEEE Statistical Signal Processing Workshop (SSP), Hanoi, Vietnam, 2–5 July 2023; IEEE: Piscataway, NJ, USA, 2023; pp. 364–368. [Google Scholar]
- Englhardt, Z.; Ma, C.; Morris, M.E.; Chang, C.C.; Xu, X.O.; Qin, L.; McDuff, D.; Liu, X.; Patel, S.; Iyer, V. From Classification to Clinical Insights: Towards Analyzing and Reasoning about Mobile and Behavioral Health Data with Large Language Models. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 2024, 8, 1–25. [Google Scholar] [CrossRef]
- Suh, S.; Rey, V.F.; Lukowicz, P. Tasked: Transformer-based adversarial learning for human activity recognition using wearable sensors via self-knowledge distillation. Knowl.-Based Syst. 2023, 260, 110143. [Google Scholar] [CrossRef]
- Alharbi, F.; Ouarbya, L.; Ward, J.A. Comparing sampling strategies for tackling imbalanced data in human activity recognition. Sensors 2022, 22, 1373. [Google Scholar] [CrossRef]
- L’Hommedieu, M.; L’Hommedieu, J.; Begay, C.; Schenone, A.; Dimitropoulou, L.; Margolin, G.; Falk, T.; Ferrara, E.; Lerman, K.; Narayanan, S.; et al. Lessons learned: Recommendations for implementing a longitudinal study using wearable and environmental sensors in a health care organization. JMIR mHealth uHealth 2019, 7, e13305. [Google Scholar] [CrossRef] [PubMed]
- Yan, S.; Hosseinmardi, H.; Kao, H.T.; Narayanan, S.; Lerman, K.; Ferrara, E. Affect estimation with wearable sensors. J. Healthc. Inform. Res. 2020, 4, 261–294. [Google Scholar] [CrossRef] [PubMed]
- Wu, R.; Yu, C.; Pan, X.; Liu, Y.; Zhang, N.; Fu, Y.; Wang, Y.; Zheng, Z.; Chen, L.; Jiang, Q.; et al. MindShift: Leveraging Large Language Models for Mental-States-Based Problematic Smartphone Use Intervention. In Proceedings of the CHI Conference on Human Factors in Computing Systems, Oahu, HL, USA, 11–16 May 2024; pp. 1–24. [Google Scholar]
- Gupta, S. Deep learning based human activity recognition (HAR) using wearable sensor data. Int. J. Inf. Manag. Data Insights 2021, 1, 100046. [Google Scholar] [CrossRef]
- Bouchabou, D.; Nguyen, S.M.; Lohr, C.; LeDuc, B.; Kanellos, I. Using language model to bootstrap human activity recognition ambient sensors based in smart homes. Electronics 2021, 10, 2498. [Google Scholar] [CrossRef]
- Kaneko, H.; Inoue, S. Toward pioneering sensors and features using large language models in human activity recognition. In Proceedings of the 2023 ACM International Joint Conference on Pervasive and Ubiquitous Computing & the 2023 ACM International Symposium onWearable Computing, Cancun, Mexico, 8–12 October 2023; pp. 475–479. [Google Scholar]
- Leng, Z.; Bhattacharjee, A.; Rajasekhar, H.; Zhang, L.; Bruda, E.; Kwon, H.; Plötz, T. Imugpt 2.0: Language-based cross modality transfer for sensor-based human activity recognition. arXiv 2024, arXiv:2402.01049. [Google Scholar]
- Civitarese, G.; Fiori, M.; Choudhary, P.; Bettini, C. Large Language Models are Zero-Shot Recognizers for Activities of Daily Living. arXiv 2024, arXiv:2407.01238. [Google Scholar]
- Liu, X.; McDuff, D.; Kovacs, G.; Galatzer-Levy, I.; Sunshine, J.; Zhan, J.; Poh, M.Z.; Liao, S.; Di Achille, P.; Patel, S. Large language models are few-shot health learners. arXiv 2023, arXiv:2305.15525. [Google Scholar]
- Shastry, K.A.; Shastry, A. An integrated deep learning and natural language processing approach for continuous remote monitoring in digital health. Decis. Anal. J. 2023, 8, 100301. [Google Scholar] [CrossRef]
- Liu, Z.; Chen, C.; Cao, J.; Pan, M.; Liu, J.; Li, N.; Miao, F.; Li, Y. Large Language Models for Cuffless Blood Pressure Measurement From Wearable Biosignals. arXiv 2024, arXiv:2406.18069. [Google Scholar]
- Das Swain, V.; Saha, K. Teacher, Trainer, Counsel, Spy: How Generative AI can Bridge or Widen the Gaps in Worker-Centric Digital Phenotyping of Wellbeing. In Proceedings of the 3rd Annual Meeting of the Symposium on Human-Computer Interaction for Work, Newcastle upon Tyne, UK, 25–27 June 2024; pp. 1–13. [Google Scholar]
- Dongre, P. Physiology-Driven Empathic Large Language Models (EmLLMs) for Mental Health Support. In Proceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems, Honolulu, HI, USA, 11–16 May 2024; pp. 1–5. [Google Scholar]
- Hegde, N.; Vardhan, M.; Nathani, D.; Rosenzweig, E.; Speed, C.; Karthikesalingam, A.; Seneviratne, M. Infusing behavior science into large language models for activity coaching. PLoS Digit. Health 2024, 3, e0000431. [Google Scholar] [CrossRef] [PubMed]
- Ragavan, V. Automated Health Coaching: A Study On Integrating Wearable Data with Large Language Models. Ph.D. Thesis, University of Illinois at Chicago, Chicago, IL, USA, 2024. [Google Scholar]
- Chiras, M. Exploration of Different Large Language Models for Retrieval-Augmented Generation in Analyzing Wearable Running Data for Sports Physiotherapy. Bachelor’s Thesis, University of Twente, Enschede, The Netherlands, 2024. [Google Scholar]
- Stefana, E.; Marciano, F.; Rossi, D.; Cocca, P.; Tomasoni, G. Wearable devices for ergonomics: A systematic literature review. Sensors 2021, 21, 777. [Google Scholar] [CrossRef]
- Patel, V.; Chesmore, A.; Legner, C.M.; Pandey, S. Trends in workplace wearable technologies and connected-worker solutions for next-generation occupational safety, health, and productivity. Adv. Intell. Syst. 2022, 4, 2100099. [Google Scholar] [CrossRef]
- Mortezapour, A. Ergonomic LLM or LLM for Ergonomics? Prompt Engineering Insights for an Interventional Case Study. Ph.D. Thesis, Hamadan University of Medical Sciences, Hamadan, Iran, 2024. [Google Scholar]
- Vaswani, A.; Shazeer, N.; Parmar, N.; Uszkoreit, J.; Jones, L.; Gomez, A.N.; Kaiser, Ł.; Polosukhin, I. Attention is all you need. In Proceedings of the 31st Conference on Neural Information Processing Systems (NIPS 2017), Long Beach, CA, USA, 4–9 December 2017; Volume 30. [Google Scholar]
- Brown, T.; Mann, B.; Ryder, N.; Subbiah, M.; Kaplan, J.D.; Dhariwal, P.; Neelakantan, A.; Shyam, P.; Sastry, G.; Askell, A.; et al. Language models are few-shot learners. In Proceedings of the Advances in Neural Information Processing Systems 33 (NeurIPS 2020), Virtual, 6–12 December 2020; Volume 33, pp. 1877–1901. [Google Scholar]
- Touvron, H.; Lavril, T.; Izacard, G.; Martinet, X.; Lachaux, M.A.; Lacroix, T.; Rozière, B.; Goyal, N.; Hambro, E.; Azhar, F.; et al. Llama: Open and efficient foundation language models. arXiv 2023, arXiv:2302.13971. [Google Scholar]
- Touvron, H.; Martin, L.; Stone, K.; Albert, P.; Almahairi, A.; Babaei, Y.; Bashlykov, N.; Batra, S.; Bhargava, P.; Bhosale, S.; et al. Llama 2: Open foundation and fine-tuned chat models. arXiv 2023, arXiv:2307.09288. [Google Scholar]
- Hota, A.; Chatterjee, S.; Chakraborty, S. Evaluating Large Language Models as Virtual Annotators for Time-series Physical Sensing Data. arXiv 2024, arXiv:2403.01133. [Google Scholar]
- Ferrara, E. Should ChatGPT be Biased? Challenges and Risks of Bias in Large Language Models. First Monday 2023, 28. [Google Scholar] [CrossRef]
- Kaseris, M.; Kostavelis, I.; Malassiotis, S. A Comprehensive Survey on Deep Learning Methods in Human Activity Recognition. Mach. Learn. Knowl. Extr. 2024, 6, 842–876. [Google Scholar] [CrossRef]
- Athota, R.K.; Sumathi, D. Human activity recognition based on hybrid learning algorithm for wearable sensor data. Meas. Sensors 2022, 24, 100512. [Google Scholar] [CrossRef]
- Augustinov, G.; Nisar, M.A.; Li, F.; Tabatabaei, A.; Grzegorzek, M.; Sohrabi, K.; Fudickar, S. Transformer-based recognition of activities of daily living from wearable sensor data. In Proceedings of the 7th International Workshop on Sensor-Based Activity Recognition and Artificial Intelligence, Rostock, Germany, 19–20 September 2022. [Google Scholar]
- Hosseinmardi, H.; Ghasemian, A.; Narayanan, S.; Lerman, K.; Ferrara, E. Tensor Embedding: A Supervised Framework for Human Behavioral Data Mining and Prediction. In Proceedings of the ICHI 2023—11th IEEE International Conference on Healthcare Informatics, Houston, TX, USA, 26–29 June 2023. [Google Scholar]
- Maharana, K.; Mondal, S.; Nemade, B. A review: Data pre-processing and data augmentation techniques. Glob. Transit. Proc. 2022, 3, 91–99. [Google Scholar] [CrossRef]
- Um, T.T.; Pfister, F.M.; Pichler, D.; Endo, S.; Lang, M.; Hirche, S.; Fietzek, U.; Kulić, D. Data augmentation of wearable sensor data for parkinson’s disease monitoring using convolutional neural networks. In Proceedings of the 19th ACM International Conference on Multimodal Interaction, Glasgow, UK, 13–17 November 2017; pp. 216–220. [Google Scholar]
- Bai, G.; Chai, Z.; Ling, C.; Wang, S.; Lu, J.; Zhang, N.; Shi, T.; Yu, Z.; Zhu, M.; Zhang, Y.; et al. Beyond efficiency: A systematic survey of resource-efficient large language models. arXiv 2024, arXiv:2401.00625. [Google Scholar]
- Lv, K.; Yang, Y.; Liu, T.; Gao, Q.; Guo, Q.; Qiu, X. Full parameter fine-tuning for large language models with limited resources. arXiv 2023, arXiv:2306.09782. [Google Scholar]
- Huang, G.; Li, Y.; Jameel, S.; Long, Y.; Papanastasiou, G. From explainable to interpretable deep learning for natural language processing in healthcare: How far from reality? Comput. Struct. Biotechnol. J. 2024, 24, 362–373. [Google Scholar] [CrossRef]
- Zhao, H.; Yang, F.; Lakkaraju, H.; Du, M. Opening the black box of large language models: Two views on holistic interpretability. arXiv 2024, arXiv:2402.10688. [Google Scholar]
- Li, Y.; Du, M.; Song, R.; Wang, X.; Wang, Y. A survey on fairness in large language models. arXiv 2023, arXiv:2308.10149. [Google Scholar]
- Gallegos, I.O.; Rossi, R.A.; Barrow, J.; Tanjim, M.M.; Kim, S.; Dernoncourt, F.; Yu, T.; Zhang, R.; Ahmed, N.K. Bias and fairness in large language models: A survey. In Computational Linguistics; MIT Press: Cambridge, MA, USA, 2024; pp. 1–79. [Google Scholar]
- Webber, M.; Rojas, R.F. Human activity recognition with accelerometer and gyroscope: A data fusion approach. IEEE Sens. J. 2021, 21, 16979–16989. [Google Scholar] [CrossRef]
- Jeyakumar, J.V.; Lai, L.; Suda, N.; Srivastava, M. SenseHAR: A robust virtual activity sensor for smartphones and wearables. In Proceedings of the 17th Conference on Embedded Networked Sensor Systems, New York, NY, USA, 10–13 November 2019; pp. 15–28. [Google Scholar]
- King, R.C.; Villeneuve, E.; White, R.J.; Sherratt, R.S.; Holderbaum, W.; Harwin, W.S. Application of data fusion techniques and technologies for wearable health monitoring. Med. Eng. Phys. 2017, 42, 1–12. [Google Scholar] [CrossRef] [PubMed]
- Yadav, S.K.; Tiwari, K.; Pandey, H.M.; Akbar, S.A. A review of multimodal human activity recognition with special emphasis on classification, applications, challenges and future directions. Knowl.-Based Syst. 2021, 223, 106970. [Google Scholar] [CrossRef]
- Dong, Q.; Chen, X.; Satyanarayanan, M. Creating edge ai from cloud-based llms. In Proceedings of the 25th International Workshop on Mobile Computing Systems and Applications, San Diego, CA, USA, 28–29 February 2024; pp. 8–13. [Google Scholar]
- Basit, A.; Hussain, K.; Hanif, M.A.; Shafique, M. MedAide: Leveraging Large Language Models for On-Premise Medical Assistance on Edge Devices. arXiv 2024, arXiv:2403.00830. [Google Scholar]
- Can, Y.S.; Ersoy, C. Privacy-preserving federated deep learning for wearable IoT-based biomedical monitoring. ACM Trans. Internet Technol. (TOIT) 2021, 21, 1–17. [Google Scholar] [CrossRef]
- Liu, J.C.; Goetz, J.; Sen, S.; Tewari, A. Learning from others without sacrificing privacy: Simulation comparing centralized and federated machine learning on mobile health data. JMIR mHealth uHealth 2021, 9, e23728. [Google Scholar] [CrossRef] [PubMed]
- Mishra, A.; Saha, S.; Mishra, S.; Bagade, P. A federated learning approach for smart healthcare systems. CSI Trans. ICT 2023, 11, 39–44. [Google Scholar] [CrossRef]
- Lawrence, H.R.; Schneider, R.A.; Rubin, S.B.; Mataric, M.J.; McDuff, D.J.; Bell, M.J. The opportunities and risks of large language models in mental health. arXiv 2024, arXiv:2403.14814. [Google Scholar]
- Chung, N.C.; Dyer, G.; Brocki, L. Challenges of large language models for mental health counseling. arXiv 2023, arXiv:2311.13857. [Google Scholar]
- Ji, S.; Zhang, T.; Yang, K.; Ananiadou, S.; Cambria, E. Rethinking large language models in mental health applications. arXiv 2023, arXiv:2311.11267. [Google Scholar]
- Kovoor, M.; Durairaj, M.; Karyakarte, M.S.; Hussain, M.Z.; Ashraf, M.; Maguluri, L.P. Sensor-enhanced wearables and automated analytics for injury prevention in sports. Meas. Sens. 2024, 32, 101054. [Google Scholar] [CrossRef]
- Seshadri, D.R.; Thom, M.L.; Harlow, E.R.; Gabbett, T.J.; Geletka, B.J.; Hsu, J.J.; Drummond, C.K.; Phelan, D.M.; Voos, J.E. Wearable technology and analytics as a complementary toolkit to optimize workload and to reduce injury burden. Front. Sport. Act. Living 2021, 2, 630576. [Google Scholar] [CrossRef]
- Chidambaram, S.; Maheswaran, Y.; Patel, K.; Sounderajah, V.; Hashimoto, D.A.; Seastedt, K.P.; McGregor, A.H.; Markar, S.R.; Darzi, A. Using artificial intelligence-enhanced sensing and wearable technology in sports medicine and performance optimisation. Sensors 2022, 22, 6920. [Google Scholar] [CrossRef] [PubMed]
- Schall, M.C., Jr.; Sesek, R.F.; Cavuoto, L.A. Barriers to the adoption of wearable sensors in the workplace: A survey of occupational safety and health professionals. Hum. Factors 2018, 60, 351–362. [Google Scholar] [CrossRef] [PubMed]
- Donisi, L.; Cesarelli, G.; Pisani, N.; Ponsiglione, A.M.; Ricciardi, C.; Capodaglio, E. Wearable sensors and artificial intelligence for physical ergonomics: A systematic review of literature. Diagnostics 2022, 12, 3048. [Google Scholar] [CrossRef] [PubMed]
- Lind, C.M.; Abtahi, F.; Forsman, M. Wearable motion capture devices for the prevention of work-related musculoskeletal disorders in ergonomics—An overview of current applications, challenges, and future opportunities. Sensors 2023, 23, 4259. [Google Scholar] [CrossRef] [PubMed]
- Vega-Barbas, M.; Diaz-Olivares, J.A.; Lu, K.; Forsman, M.; Seoane, F.; Abtahi, F. P-Ergonomics Platform: Toward precise, pervasive, and personalized ergonomics using wearable sensors and edge computing. Sensors 2019, 19, 1225. [Google Scholar] [CrossRef] [PubMed]
- Ferrara, E. The Butterfly Effect in Artificial Intelligence Systems: Implications for AI Bias and Fairness. Mach. Learn. Appl. 2024, 15, 100525. [Google Scholar] [CrossRef]
- Kim, J.W.; Moon, S.-M.; Kang, S.-u.; Jang, B. Effective privacy-preserving collection of health data from a user’s wearable device. Appl. Sci. 2020, 10, 6396. [Google Scholar] [CrossRef]
- Gupta, B.B.; Gaurav, A.; Arya, V.; Alhalabi, W.; Alsalman, D.; Vijayakumar, P. Enhancing user prompt confidentiality in Large Language Models through advanced differential encryption. Comput. Electr. Eng. 2024, 116, 109215. [Google Scholar] [CrossRef]
- Raeini, M. Privacy-Preserving Large Language Models (PPLLMs). SSRN 4512071 2023. Available online: https://www.researchgate.net/publication/372950300_Privacy-Preserving_Large_Language_Models_PPLLMs (accessed on 9 July 2024).
- Yan, B.; Li, K.; Xu, M.; Dong, Y.; Zhang, Y.; Ren, Z.; Cheng, X. On protecting the data privacy of large language models (llms): A survey. arXiv 2024, arXiv:2403.05156. [Google Scholar]
- Rasnayaka, S.; Sim, T. Your tattletale gait privacy invasiveness of IMU gait data. In Proceedings of the 2020 IEEE International Joint Conference on Biometrics (IJCB), Houston, TX, USA, 28 September–1 October 2020; IEEE: Piscataway, NJ, USA, 2020; pp. 1–10. [Google Scholar]
- Duan, H.; Dziedzic, A.; Yaghini, M.; Papernot, N.; Boenisch, F. On the privacy risk of in-context learning. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, Toronto, ON, Canada, 9–14 July 2023. [Google Scholar]
- Sebastian, G. Privacy and data protection in ChatGPT and other AI Chatbots: Strategies for securing user information. Int. J. Secur. Priv. Pervasive Comput. (IJSPPC) 2023, 15, 1–14. [Google Scholar] [CrossRef]
- Banerjee, A.; Gupta, S.K.; Venkatasubramanian, K.K. PEES: Physiology-based end-to-end security for mhealth. In Proceedings of the 4th Conference on Wireless Health, Baltimore, MD, USA, 1–3 November 2013; pp. 1–8. [Google Scholar]
- Jan, M.A.; Khan, F.; Mastorakis, S.; Adil, M.; Akbar, A.; Stergiou, N. LightIoT: Lightweight and secure communication for energy-efficient IoT in health informatics. IEEE Trans. Green Commun. Netw. 2021, 5, 1202–1211. [Google Scholar] [CrossRef] [PubMed]
- Yan, S.; Kao, H.t.; Ferrara, E. Fair class balancing: Enhancing model fairness without observing sensitive attributes. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management, Virtual Event, 19–23 October 2020; pp. 1715–1724. [Google Scholar]
- Erfani, A.; Frias-Martinez, V. A fairness assessment of mobility-based COVID-19 case prediction models. PLoS ONE 2023, 18, e0292090. [Google Scholar] [CrossRef] [PubMed]
- Yfantidou, S.; Sermpezis, P.; Vakali, A.; Baeza-Yates, R. Uncovering Bias in Personal Informatics. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 2023, 7, 1–30. [Google Scholar] [CrossRef]
- Alam, M.A.U. Ai-fairness towards activity recognition of older adults. In Proceedings of the MobiQuitous 2020-17th EAI International Conference on Mobile and Ubiquitous Systems: Computing, Networking and Services, Darmstadt, Germany, 7–9 December 2020; pp. 108–117. [Google Scholar]
- Bharti, P.; Dey, A.K.; Chellappan, S.; Beckie, T. An Experimental Investigation Comparing Age-Specific and Mixed-Age Models for Wearable Assisted Activity Recognition in Women. In Proceedings of the 12th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2019), Prague, Czech Republic, 22–24 February 2019; pp. 367–374. [Google Scholar]
- Alday, E.A.P.; Rad, A.B.; Reyna, M.A.; Sadr, N.; Gu, A.; Li, Q.; Dumitru, M.; Xue, J.; Albert, D.; Sameni, R.; et al. Age, sex and race bias in automated arrhythmia detectors. J. Electrocardiol. 2022, 74, 5–9. [Google Scholar] [CrossRef]
- Zanna, K.; Sridhar, K.; Yu, H.; Sano, A. Bias reducing multitask learning on mental health prediction. In Proceedings of the 2022 10th International Conference on Affective Computing and Intelligent Interaction (ACII), Nara, Japan, 17–21 October 2022; IEEE: Piscataway, NJ, USA, 2022; pp. 1–8. [Google Scholar]
- Li, F.; Wu, P.; Ong, H.H.; Peterson, J.F.; Wei, W.Q.; Zhao, J. Evaluating and mitigating bias in machine learning models for cardiovascular disease prediction. J. Biomed. Inform. 2023, 138, 104294. [Google Scholar] [CrossRef]
- Kruspe, A. Towards detecting unanticipated bias in Large Language Models. arXiv 2024, arXiv:2404.02650. [Google Scholar]
- Eigner, E.; Händler, T. Determinants of llm-assisted decision-making. arXiv 2024, arXiv:2402.17385. [Google Scholar]
- Liao, Q.V.; Vaughan, J.W. Ai transparency in the age of llms: A human-centered research roadmap. arXiv 2023, arXiv:2306.01941. [Google Scholar] [CrossRef]
- Dirgová Luptáková, I.; Kubovčík, M.; Pospíchal, J. Wearable sensor-based human activity recognition with transformer model. Sensors 2022, 22, 1911. [Google Scholar] [CrossRef]
- Ferrara, E. GenAI Against Humanity: Nefarious Applications of Generative Artificial Intelligence and Large Language Models. J. Comput. Soc. Sci. 2024, 7, 549–569. [Google Scholar] [CrossRef]
- Ezzeldin, Y.H.; Yan, S.; He, C.; Ferrara, E.; Avestimehr, S. Fairfed: Enabling Group Fairness in Federated Learning. In Proceedings of the AAAI 2023—37th AAAI Conference on Artificial Intelligence, Washington, DC, USA, 7–14 February 2023. [Google Scholar]
- Wang, Y.; Xu, H.; Liu, Y.; Wang, M.; Wang, Y.; Yang, Y.; Zhou, S.; Zeng, J.; Xu, J.; Li, S.; et al. A novel deep multifeature extraction framework based on attention mechanism using wearable sensor data for human activity recognition. IEEE Sens. J. 2023, 23, 7188–7198. [Google Scholar] [CrossRef]
- Zhang, Y.; Wang, L.; Chen, H.; Tian, A.; Zhou, S.; Guo, Y. IF-ConvTransformer: A framework for human activity recognition using IMU fusion and ConvTransformer. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 2022, 6, 1–26. [Google Scholar] [CrossRef]
- Semwal, V.B.; Gupta, A.; Lalwani, P. An optimized hybrid deep learning model using ensemble learning approach for human walking activities recognition. J. Supercomput. 2021, 77, 12256–12279. [Google Scholar] [CrossRef]
- Sarkar, A.; Hossain, S.S.; Sarkar, R. Human activity recognition from sensor data using spatial attention-aided CNN with genetic algorithm. Neural Comput. Appl. 2023, 35, 5165–5191. [Google Scholar] [CrossRef]
Sensor Type | Description | Refs. |
---|---|---|
Physiological Sensors | Monitor vital signs and other physiological parameters. Examples include heart rate monitors, electrocardiograms (ECG), blood pressure monitors, and pulse oximeters. | [14] |
Motion Sensors | Include accelerometers, gyroscopes, and magnetometers, used to track movement and orientation. Essential in applications like activity recognition and sports science. | [15] |
Environmental Sensors | Detect environmental conditions such as temperature, humidity, and light. Often integrated into wearable devices to provide context-aware services. | [16] |
Biochemical Sensors | Measure biochemical markers such as glucose levels, lactate, and electrolytes. Valuable in medical diagnostics and continuous health monitoring. | [3] |
Multisensor Systems | Integrate multiple sensor types into a single device to provide comprehensive monitoring capabilities. Examples include smartwatches and fitness trackers. | [4] |
Data Type | Description | Refs. |
---|---|---|
Time-Series Data | Most wearable sensors produce continuous streams of time-series data, capturing dynamic changes over time. These types of data require specialized techniques for preprocessing, segmentation, and feature extraction. | [12] |
Multimodal Data | Wearable devices often generate multimodal data by combining inputs from different types of sensors. For instance, a smartwatch may collect both motion and physiological data simultaneously. Integrating and synchronizing these data streams is a complex task that is essential for accurate analysis. | [4] |
High-Dimensional Data | The raw data from wearable sensors can be high-dimensional, particularly when multiple sensors are used. Dimensionality reduction techniques, such as principal component analysis (PCA) and feature selection methods, are employed to manage this complexity. | [17] |
Noisy and Incomplete Data | Wearable sensors are prone to generating noisy and sometimes incomplete data due to various factors like sensor malfunctions, user movement, and environmental interference. Effective data cleaning and imputation methods are critical in maintaining data quality. | [18] |
Aspect | Description | Refs. |
---|---|---|
Capabilities of LLMs | ||
Natural Language Understanding | LLMs can comprehend complex language patterns, making them suitable for tasks such as sentiment analysis, entity recognition, and language translation. Their ability to understand context and generate relevant responses has been demonstrated in various studies, including those focusing on health data interpretation. | [3,42] |
Text Generation | The text generation capabilities of LLMs are unparalleled, allowing them to produce coherent and contextually appropriate text for diverse applications. This has been leveraged in generating health-related content, educational materials, and even creative writing. | [39] |
Multimodal Data Integration | LLMs can be integrated with other data modalities, such as sensor data, to provide comprehensive insights. For example, the LLaSA model combines text data with inertial measurement unit (IMU) data to enhance human activity recognition. | [4] |
Limitations of LLMs | ||
Computational Requirements | Training and deploying LLMs requires substantial computational resources, which can be prohibitive for many applications. Efficient model architectures and optimization techniques are necessary to mitigate these challenges. | [15] |
Data Dependency | LLMs rely heavily on large, high-quality datasets for training. The quality and diversity of the training data significantly impact the model’s performance and generalizability. Incomplete or biased data can lead to inaccurate predictions and outputs. | [12,43] |
Interpretability | LLMs operate as black-box models, making it difficult to interpret their decision-making processes. This lack of transparency is a significant limitation, especially in critical applications such as healthcare, where understanding the rationale behind predictions is crucial. | [13] |
Ethical Concerns | The use of LLMs raises ethical issues related to data privacy, security, and potential misuse. Ensuring compliance with data protection regulations and implementing privacy-preserving techniques are essential to address these concerns. | [44] |
Challenge | Description | Refs. |
---|---|---|
General Issues Related to Using LLMs with Wearable Sensor Data | ||
Data Quality and Preprocessing | Wearable sensors generate noisy, incomplete, and inconsistent data. Techniques like noise reduction, normalization, feature extraction, hybrid sampling, and data augmentation are essential for transforming raw data into a suitable format for LLMs. Effective multimodal data integration is also critical. | [1,2,48,49] |
Computational Requirements | LLMs require substantial computational resources for training and fine-tuning, impacting their feasibility for real-time data analysis. Optimizing the model architectures, training algorithms, and integration with other AI techniques can help to reduce the resource requirements. | [50,51] |
Interpretability and Transparency | LLMs operate as black-box models, making it difficult to understand their decision-making processes. Improving their interpretability through attention mechanisms, visualization, and explainable AI techniques is crucial for trust and accountability, especially in healthcare applications. | [52,53] |
Bias and Fairness | LLMs may exhibit biases based on the demographic characteristics of the training data. Ensuring fairness involves using diverse datasets, regularly auditing model performance, and implementing bias mitigation strategies. | [43,54,55] |
Challenges Specific to Using LLMs for Sensor-Based HAR | ||
Data Processing and Integration Complexities | Sensor-based HAR involves harmonizing different data types from various wearable devices. Developing robust data fusion techniques and ensuring the synchronization and alignment of multimodal data are essential. | [56,57,58,59] |
Real-Time Adaptability | LLMs may struggle with low-latency requirements for real-time applications like emergency response and health monitoring. Exploring lightweight model architectures (or tiny-LLMs), efficient training algorithms, and edge computing can help to achieve real-time adaptability. | [60,61] |
Challenges Specific to Using LLMs for Sensor-Based Health Monitoring and Mental Health | ||
Health Monitoring | The continuous tracking of physiological metrics using wearable sensors and LLMs requires accurate health predictions. Techniques like federated learning can enhance the accuracy and privacy in health monitoring systems. | [62,63,64] |
Mental Health | LLMs can provide personalized mental health support but require stringent privacy protections and timely, contextually appropriate interventions. Models like EmLLMs aim to provide empathetic and personalized support. | [31,65,66,67] |
Challenges Specific to Using LLMs for Sensor-Based Behavioral Modeling in Sports and Ergonomics | ||
Sports and Fitness | Wearable sensors are used to monitor performance, optimize training, and prevent injuries. Ensuring accurate and relevant insights through multimodal data integration and real-time analytics is critical for LLMs. | [32,68,69,70] |
Workplace Ergonomics | Wearable sensors can improve workplace ergonomics by monitoring movements and posture, providing real-time feedback, and preventing musculoskeletal disorders. Ensuring data accuracy, personalized recommendations, and managing privacy are key challenges for LLMs. | [37,71,72,73,74] |
Ethical or Legal Issue | Description | Refs. |
---|---|---|
User Privacy Protection | ||
General Privacy Concerns | Wearable devices collect extensive personal data, including health metrics, location information, and behavioral patterns. Protecting user privacy is paramount to maintain trust and ensure confidentiality. | [14,76,77,78,79] |
Sensor-Specific Privacy Concerns | Privacy concerns vary with the type of sensor data (e.g., IMU, ECG, PPG). IMU data might pose a lower privacy risk compared to ECG and PPG data, which are more sensitive and related to health conditions. Stronger privacy protections are necessary for more sensitive data types processed via LLMs. | [79,80,81] |
Data Handling Practices | Implementing data anonymization, minimization techniques, and transparent data handling practices is essential while using LLMs with sensor-based data. | [77,82] |
Data Security | ||
General Data Security Concerns | Robust encryption methods, secure data storage solutions, and access control mechanisms are essential to safeguard wearable sensor data from unauthorized access and cyberattacks. Compliance with GDPR, HIPAA, and other regulations is necessary. | [21,83,84] |
Privacy-Preserving Techniques | Techniques like federated learning and differential privacy minimize the risk of data exposure while enabling the effective use of LLMs for data analysis. These techniques keep raw data on the user’s device, sharing only aggregated model updates. | [13,85] |
Sensor-Specific Data Security | Tailored security measures for different sensor types (e.g., ECG and PPG data require stronger encryption than IMU data) are necessary. Ensuring secure transmission from wearable devices to LLM-hosting infrastructure or edge computing nodes is also crucial. | [82,83,84] |
Model Bias | ||
General Model Bias Concerns | LLMs are susceptible to biases in their training data, leading to unfair and discriminatory outcomes. Diverse and representative training datasets, regular audits, and bias correction techniques are essential. | [43,54,75] |
Sensor-Specific Model Bias | The bias can vary with different sensor types (e.g., IMU vs. ECG) and demographic characteristics (e.g., young vs. elderly users). The continuous monitoring and adjustment of models are required to ensure fair performance across all user groups. | [86,87,88,89,90,91,92] |
Transparency in Model Development | Ensuring transparency in sensor-based LLM model development and decision-making processes helps to identify and address biases effectively. Clear communication regarding how models work and how decisions are made is crucial. | [93,94,95] |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2024 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Ferrara, E. Large Language Models for Wearable Sensor-Based Human Activity Recognition, Health Monitoring, and Behavioral Modeling: A Survey of Early Trends, Datasets, and Challenges. Sensors 2024, 24, 5045. https://doi.org/10.3390/s24155045
Ferrara E. Large Language Models for Wearable Sensor-Based Human Activity Recognition, Health Monitoring, and Behavioral Modeling: A Survey of Early Trends, Datasets, and Challenges. Sensors. 2024; 24(15):5045. https://doi.org/10.3390/s24155045
Chicago/Turabian StyleFerrara, Emilio. 2024. "Large Language Models for Wearable Sensor-Based Human Activity Recognition, Health Monitoring, and Behavioral Modeling: A Survey of Early Trends, Datasets, and Challenges" Sensors 24, no. 15: 5045. https://doi.org/10.3390/s24155045
APA StyleFerrara, E. (2024). Large Language Models for Wearable Sensor-Based Human Activity Recognition, Health Monitoring, and Behavioral Modeling: A Survey of Early Trends, Datasets, and Challenges. Sensors, 24(15), 5045. https://doi.org/10.3390/s24155045