Predictive Model of Psychological Factors Influencing University Students' Intentions and Behaviour Regarding Physical Exercise—Based on Machine Learning Algorithms

Insufficient physical exercise has become a significant risk factor affecting university students' physical and mental health, with participation behaviour influenced by the complex interplay of multidimensional psychological and physiological variables. This study aims to utilise machine learning algorithms to construct predictive models for university students' physical exercise participation behaviour, identify key influencing factors, and validate the diversity gradient hypothesis. Based on questionnaire survey data from university students across multiple provinces nationwide, five machine learning algorithms—Multi-Layer Perceptron (MLP), XGBoost, Random Forest, Gradient Boosting Decision Tree (GBDT), and Decision Tree—were employed for modelling and comparative analysis. Results indicate that among the five models, GBDT exhibits optimal classification performance. Feature importance analysis identifies weight and anxiety as core variables influencing model predictions. Propensity score propensity score maps further reveal marginal effects of different feature value ranges on predictions, particularly evident in assessing tendencies towards high-intensity exercise. The decision tree model features resistance to temptation as its root node, with healthy habits and impulse control as sub-nodes. Integrating the Multi-Process Action Control (M-PAC) model with stress process theory, this study elucidates at the mechanistic level how the interplay between physical and mental health, self-control capacity, and emotion regulation strategies collectively influences physical exercise behaviour. It proposes that systematic health intervention and educational planning strategies can promote the development of exercise habits among university students.