Machine learning‐based early warning model for adolescent mental health risk using the p factor

BACKGROUND: Accurate identification of adolescents at high mental health risk is essential for timely intervention and improved service delivery. The general factor of psychopathology (p factor), capturing shared variance across mental disorders, was used as a transdiagnostic outcome to develop a machine learning model for risk identification. METHODS: Data from 5,283 Chinese adolescents were used for model training and testing, with an additional 968 participants for external validation. Guided by ecological systems theory, a multidimensional early warning framework covering 59 indicators across individual, school, family, and society domains was constructed. Seven machine learning algorithms were tested using the p factor as the outcome. Shapley Additive exPlanations (SHAP) values ranked predictor importance and identified an optimal feature subset. RESULTS: The XGBoost model showed the best performance (macro F1 = 0.73 on internal validation, 0.80 on external validation). The final model used 23 predictors. SHAP confirmed sleep quality as the most influential factor, followed by repetitive negative thinking, interpersonal stress, impulsivity, and emotional intensity. CONCLUSIONS: This study suggests that a p-factor-based machine learning model can effectively identify adolescents at mental health risk, with sleep quality emerging as the most influential predictor, suggesting potential intervention targets for early prevention.