AI-Coached Fitness Tracking for Personalized Health Monitoring

A federated fitness monitoring future composed of AI-assisted personalized fitness coach, fitness sensor that can be worn, fitness activity analysis (personalized fitness coaching suggestion), recovery prediction and injury-risk assessment are proposed in this paper. With 12 weeks of sessions to study, it was a controlled synthetic data, with 1800 sessions records of 150 simulated participants. The variables included user profile data, data from sensors which detect movement, physiology data, findings of training-load and recovery indicators and coaching labels. Two baseline models, one more conventional, and the other simpler, were developed for SOOT. The test accuracy and macro-F1 score of the activity-recognition model (0.9318 and 0.9317, respectively) and coaching-recommendation model (0.9495 and 0.8968, respectively) supported that their models can be utilized successfully. Based on the results, it is concluded that structured sensor and recovery can be used to create a prototype level fitness advice and audibility of the advice rules is reliable. Clinical readiness has nothing to do with them - just a by-product of relevance. Systematically designs and prototypes. Passive tracking" is to report and detail basically something that happened. The recommendation on whether the user should keep doing the same action, take an alternate course of action or stop should be provided when there is uncertainty in the AI system, user fatigue, and adherence to the AI system. The dependency between the activity-labels and the healthstate-variables are not hidden or camouflaged in the proposed framework they are presented as a means to express, validate and to document this dependency. Future studies will involve consent based data, as well as with all other types of data being collected with wearables, across different brands with different age groups, training history and training environments. The following should be included in the Model evaluation: Calibration Error, Fairness Tests, Sensor Drift Analysis, Longitudinal Adherence and Adverse response monitoring. Maybe some strong temporal (weekly) models, such as recurrent models, and transformer models could be improved in this timeframe (weekly), but should be explainable. Additionally, a practical system should contain functionality to take into consideration reason codes and confidence/safety limitations relating to each of the recommendations. With none of those checks and balances an other mysterious notification system would be an AI coach. As long as they have, it can be a physical fitness managing support layer which can be measurable! The next step of research is to compare automated coaching to human coaching (and compare the mix of measuring techniques to determine outcomes and how to report error rates, pre-registered outcomes and transparent error reporting).

[1]     A. Hassoon et al., “Randomized trial of two artificial intelligence coaching interventions to increase physical activity in cancer survivors,” npj Digital Medicine, vol. 4, no. 1, Art. no. 168, Dec. 2021, doi: 10.1038/s41746-021-00539-9. [2]     F. Yang, C. Wang, M. A. Lindsey, and R. An, “Use of Artificial-Intelligence-Based Chatbots to Promote Physical Activity: A Systematic Review of Interventions,” Kinesiology Review, vol. 14, no. 1, pp. 80–92, Feb. 2025, doi: 10.1123/kr.2024-0081. [3]     C. Loughnane, J. Laiti, R. O’Donovan, and P. J. Dunne, “Systematic review exploring human, AI, and hybrid health coaching in digital health interventions: Trends, engagement, and lifestyle outcomes,” Frontiers in Digital Health, vol. 7, Art. no. 1536416, 2025, doi: 10.3389/fdgth.2025.1536416. [4]     A. Montelaghi, A. M. Ciorciari, R. Roklicer, G. Jurak, and A. Carraro, “AI in motion: A systematic review of artificial intelligence-driven virtual assistants for physical activity promotion and their comparison with traditional strategies,” Artificial Intelligence Review, vol. 58, Art. no. 367, 2025, doi: 10.1007/s10462-025-11361-0. [5]     E. Gabarron, D. Larbi, O. Rivera-Romero, and K. Denecke, “Human Factors in AI-Driven Digital Solutions for Increasing Physical Activity: Scoping Review,” JMIR Human Factors, vol. 11, Art. no. e55964, 2024, doi: 10.2196/55964. [6]     A. Brons, S. Wang, B. Visser, B. Kröse, S. Bakkes, and R. Veltkamp, “Machine Learning Methods to Personalize Persuasive Strategies in mHealth Interventions That Promote Physical Activity: Scoping Review and Categorization Overview,” Journal of Medical Internet Research, vol. 26, Art. no. e47774, 2024, doi: 10.2196/47774. [7]     J. Lopez-Barreiro, J. L. Garcia-Soidan, L. Alvarez-Sabucedo, and J. M. Santos-Gago, “Artificial Intelligence-Powered Recommender Systems for Promoting Healthy Habits and Active Aging: A Systematic Review,” Applied Sciences, vol. 14, no. 22, Art. no. 10220, 2024, doi: 10.3390/app142210220. [8]     H. Jubair and M. Mehenaz, “Utilizing machine learning algorithms for personalized workout recommendations and monitoring: A systematic review on smartwatch-assisted exercise prescription,” Digital Health, vol. 11, 2025, doi: 10.1177/20552076251355365. [9]     A. Olyanasab and M. Annabestani, “Leveraging Machine Learning for Personalized Wearable Biomedical Devices: A Review,” Journal of Personalized Medicine, vol. 14, no. 2, Art. no. 203, 2024, doi: 10.3390/jpm14020203. [10]   R. Li et al., “Wearable Activity Tracker–Based Interventions for Physical Activity, Body Composition, and Physical Function Among Community-Dwelling Older Adults: Systematic Review and Meta-Analysis of Randomized Controlled Trials,” Journal of Medical Internet Research, vol. 27, Art. no. e59507, 2025, doi: 10.2196/59507.