Association Between New Tea Drink Consumption and Mental Health Disorders Among Chinese College Students: An Interpretable Machine Learning Approach.

BACKGROUND: With rapid lifestyle changes, mental health disorders and chronic diseases are increasingly prevalent among college students, driven by multiple risk factors. Given the limitations of traditional linear models in managing high-dimensional data, this study employed machine learning to predict these risks. METHODS: A cross-sectional survey was conducted among 1217 college students across 14 provinces in China. Information on new tea drink consumption, psychological traits, and demographic characteristics was collected. Five machine learning algorithms, Random Forest, XGBoost, Logistic Regression, Support Vector Machine, and Artificial Neural Network were evaluated. RESULTS: The XGBoost model achieved robust performance for mental health disorders (AUC = 0.89, Accuracy = 0.86), but showed limited sensitivity for chronic diseases due to class imbalance. SHapley Additive exPlanations analysis indicated that insomnia symptoms were the primary predictor for mental health disorders. Mediation analysis suggested that the total effect between new tea drink frequency and mental health disorders was 0.254, and insomnia symptoms partially mediated this association (effect = 0.062; share = 24.4%). CONCLUSION: These findings suggest that integrating machine learning into early screening systems may help inform preliminary screening and risk-stratification efforts for college student mental health. By identifying associated behavioral patterns, these models offer a basis for hypothesis generation, though longitudinal research is necessary to establish causal relationships and validate their utility.