Predicting and Interpreting Psychological Distress and Mental Health Treatment Experience from Life Experiences and Everyday Behaviors

The aim of this study was to examine whether psychological distress and treatment experience among university students can be predicted using machine learning models based on daily experiences and behaviors. Participants were university students who completed a web-based survey assessing psychological distress (K10 scores), treatment experience, and various life and behavioral factors. To reduce self-report bias, behaviorally specific and concrete variables were selected as predictors. Multiple models, including linear/logistic regression, Random Forest, and CatBoost, were trained and evaluated using nested five-fold cross-validation. The models explained approximately 30% of the variance in K10 scores, and classification performance reached an AUC of .77. Across both studies, key predictors consistently included seeking mental health information, persistent rumination over minor issues, and perceived lack of parental respect. Additional predictors included satisfaction with university entrance exam results and drinking frequency for distress, and experiences of bullying and family history of mental illness for treatment experience. These findings suggest that everyday experiences and behaviors may serve as useful indicators of psychological distress and help-seeking among university students.