Adaptive Optimization of Personalized Exercise Regimens Using Proximal Policy Optimization and Data Mining

Abstract

Regular physical activity is essential for maintaining good health; however, people have varying fitness levels, health conditions, and goals. Most traditional exercise programs employ a one-size-fits-all approach, often resulting in suboptimal results or a lack of motivation. A personalized approach is necessary to better match individual needs and help users stay on track. Advanced technology and machine learning enable the collection of detailed activity data and the development of intelligent training systems. This paper proposes an intelligent method to create personalized exercise programs using Deep Reinforcement Learning (DRL) and data mining (PEP-DRL-DM). The system utilizes the PAMAP2 Physical Activity Monitoring dataset, which comprises sensor data such as heart rate and movement during various activities. Data mining techniques are applied to learn patterns from this data, such as user fitness levels, activity habits, and performance trends. These patterns help the DRL model understand each user’s current state. Then, Proximal Policy Optimization (PPO) is used to decide the best type, duration, and intensity of exercises. A virtual training setup gives feedback based on how users improve over time. Experimental results indicate that PEP-DRL-DM obtains a 17.11% improvement in fitness result, a personalization score of 0.65 to 0.93, and 85% user retention over 10 sessions, surpassing baseline methods. The system adapted well to different user needs and fitness conditions. In conclusion, combining data mining with PPO helps build personalized and flexible exercise programs that improve user progress and engagement over time.