Predicting Chinese adolescents' non-suicidal self-injury urges at diathetic, environmental, and life-event...

This study employed machine learning (ML) and network analysis to identify key predictors of non-suicidal self-injury (NSSI) urges among a sample of Chinese adolescents (N = 1774) in a 10-month longitudinal survey. The Stacking ensemble ML model achieved optimal prediction (AUC = 0.72). SHAP analysis revealed significant multi-level predictors including diatheses (gender and ego-depletion), environmental factors (emotional abuse and neglect), and life events (emotional relative deprivation, peer stressors, and academic stressors). Network analysis was further used to explore the interaction patterns among key predictors and identified peer stressor and ego-depletion as the central nodes in both genders, with notable structural and global strength invariance across groups. These findings offer a theoretical foundation for early identification and targeted interventions for NSSI urges.