Personalized Fitness Recommendation System Integrating Yoga, Physiotherapy Exercises and Tips

Fitness applications have been a rage in the past decade, even more so since the SARS-CoV-2 pandemic limited outdoor activities and gym trips. The applications themselves, though, are marred by low adherence and high attrition. This paper discusses the Yoga, Physiotherapy Exercises, and Health Tips Fitness application that utilizes the application of machine learning to achieve maximum user adherence and engagement. A survey of 250 users named the top three reasons for the failure of fitness applications: lack of time, generic workout plans, and rigidity. The application remedies these by using personalized workout plans based on user preference and time at hand, e.g., short workout plans and AI-based plans. We utilized Gaussian Mixture Models (GMM) for user clustering based on initial activity, and supplemented it with sophisticated techniques such as BERT4Rec for recommendation, Proximal Policy Optimization (PPO) for plan optimization, SVD for dimensionality reduction, and XGBoost for predictive user adherence analysis. The model achieved high predictive accuracy of 95.12%, a loss of 0.0349, a total loss of 0.0314, and similarly high F1 scores, establishing the model's predictive ability. The application combines yoga, physiotherapy exercises, and personalized health tips in a novel combination, addressing multiple aspects of well-being to minimize attrition and encourage frequent exercise among diverse user groups.