Machine learning-based association analysis and cross-sectional classification of physical exercise behaviors and mental health among vocational medical students

Psychological distress is common among vocational medical students, yet how multidimensional exercise behavior relates to their mental health remains coarsely described. This cross-sectional study surveyed 1,127 students at three vocational medical colleges in China between September and November 2024, applying machine learning to two analytically separate tasks: describing association patterns between exercise profiles and psychological status, and classifying concurrent psychological risk. Exercise behavior was measured with the IPAQ Short Form plus purpose-written items on modality, motivation, social context and temporal regularity; psychological status with the GAD-7, PHQ-9, PSS-10 and SWLS, combined into three ordered categories that are operational rather than diagnostic. Random forest ranking, Apriori rule mining and K-Means profiling agreed on three points: weekly frequency carried the most information among behavioral variables; frequent moderate-intensity aerobic activity co-occurred most often with low anxiety and depression; and four partially overlapping profiles emerged. XGBoost classified best on an internal hold-out test set of 226 students, one fifth of the full sample (accuracy 0.889, 95% CI 0.842 to 0.924; macro-averaged AUC 0.943), yet reached only 0.769 sensitivity for severe distress, and accuracy fell to 0.858 under leave-one-college-out validation. Because the design is cross-sectional and validation internal, these findings describe associations rather than supply a screening tool.