Evaluation of physical training effectiveness for college students based on deep reinforcement learning

Abstract

The physical training of college students can be made more productive with the help of adaptive approaches. Deep reinforcement learning (DRL) offers training policies that are more customized based on performance and physiological measurements and are superior to fixed schedules. The first goal is to make a smart system that will be efficient in physical training optimization with the help of DRL. Research suggests the Hare Escape-Search Functional Dueling Deep Q-Network (HE-FDDQN), which is a DRL algorithm that predicts student states, actions, and rewards in order to maximize training policies. Dueling architecture improves state-value learning, while escape-search enhances exploration for adaptive offline training recommendations. The College Student Physical Training Dataset was used for evaluation, which is a public data set consisting of 1000 records from the measurements of the wearable devices, fitness tests, and self-reported physical activity indicators. Preprocessing includes data cleaning, synchronization of data, and normalization of data. Python implementations exhibit effective policy learning, and visualization reveals that the development of fitness and less overtraining is possible. The HE-FDDQN model demonstrated great regression ability (MAE 1.17, RMSE 1.52, R2 0.931), and good classification scores, recall 0.922, precision 0.913, accuracy 0.925, AUC 0.912, and F1-score 0.909. The Dueling DQN strategy achieved better rewards and convergence, supporting adaptive recommendations under offline experimental evaluation. Overall, the results suggest that HE-FDDQN can enhance the assessment and personalization of college students' physical training, offering an efficient and customized alternative to fixed training plans with the potential to improve fitness outcomes.