Bayesian Neural Network for Predicting Scores of Student Physical Fitness

ABSTRACT Forecasting the physical fitness of university students holds significant importance, as fitness is not only a key health indicator but also a critical factor influencing academic performance and overall well‐being. Accurate predictions of future fitness levels can inform targeted interventions, enabling institutions to enhance student health outcomes effectively. In this work, we propose a Bayesian neural network (BNN) approach for predicting student physical fitness. The proposed method offers two key advantages, one of which is that it provides confidence measures (e.g., prediction variance) alongside forecasts, and its inherent Bayesian framework helps mitigate overfitting, improving generalization. Our experimental results demonstrate that the proposed BNN model successfully predicts nine key physical fitness indicators while also identifying the most influential factors affecting prediction accuracy. Comparative evaluations show that our method outperforms baseline approaches, achieving a prediction accuracy of 93.7%. Notably, among the nine indicators, pull‐up and sit‐up performance exhibit a substantially stronger impact on overall fitness predictions compared to the other seven indicators. These findings underscore the efficacy of our Bayesian neural network in forecasting student physical fitness, offering a robust tool for educators and health professionals to support data‐driven fitness assessments and interventions.