Behavioral and psychosocial indicators for depression and anxiety risk stratification among college students: development and external validation of a machine-learning-assisted framework

Background Depression and anxiety symptoms are common among college students. Behavioral and psychosocial indicators may support preventive campus mental health work, but risk-stratification frameworks require external validation. Methods We developed a construct-level model-explanation framework using questionnaire data from a development sample. The cleaned dataset included 1,267 students; internal-validation analyses included 1,096 students for depression risk and 1,095 for anxiety risk. Risk was defined as a Patient Health Questionnaire-9 (PHQ-9) or Generalized Anxiety Disorder-7 (GAD-7) score of 10 or higher. The primary Feature set C random-forest (RF) pipelines integrated demographic, traditional, and emerging behavioral and psychosocial indicators. Fold-specific preprocessing and percentile cutoffs were used for internal validation. Locked pipelines were applied without retraining, predictor reselection, optimization, recalibration, or threshold adjustment to an independent university sample of 1,468 students. Results External area under the receiver operating characteristic curve (AUROC), area under the precision–recall curve (PR-AUC), and Brier scores were 0.813 (95% CI, 0.776–0.848), 0.402 (0.325–0.491), and 0.090 (0.083–0.098), respectively, for depression and 0.809 (0.765–0.851), 0.205 (0.144–0.300), and 0.063 (0.056–0.070), respectively, for anxiety. Ordinary participant-level bootstrap was the primary approach, with stratified bootstrap as a sensitivity analysis. Depression calibration was close to the prevalence baseline (Brier 0.091; intercept −0.668; slope 1.401), whereas anxiety showed miscalibration (Brier 0.063 vs. baseline 0.051; intercept −1.227; slope 1.102). The models supported external discrimination and risk ranking but did not validate individual absolute-risk probabilities or clinical thresholds. Conclusion Behavioral and psychosocial indicators supported externally validated risk stratification in an independent university sample. Anxiety-risk findings should be interpreted as preliminary because the number of positive cases was limited. Further evaluation is needed before wider implementation.