Predictive Analytics for Early Detection of Youth Mental Health Risks in Underserved Schools

Abstract

Adolescent mental health issues are increasing, with many students reporting sadness and anxiety. Schools can help identify early warning signs, but limited resources often delay timely support, especially in underserved areas. This study explored predictive analytics as a school-centric approach for the early detection of youth mental health risks in underserved educational settings. Grounded in ecological and risk-and-resilience frameworks, the research synthesized existing predictive models and identified key indicators such as academic performance, bullying victimization, sleep disturbances, and substance use from publicly available youth survey and school policy datasets. A modular data architecture was proposed that integrates student-level behavioral and demographic variables with school-level contextual factors, including policy strength, counselor ratios, and climate indicators. Using synthetic data derived from publicly available youth survey and school-policy indicators, penalized logistic regression, random forest, and XGBoost models were evaluated, achieving moderate discriminatory power with AUC values ranging from 0.70 to 0.75. Fairness assessments highlighted trade-offs across racial groups, emphasizing the need for equitable deployment. Ethical, privacy, and implementation guidelines were developed to support feasible adoption in low-resource schools. Results demonstrated the value of leveraging routine school data for proactive risk stratification and targeted support. The study concludes that predictive analytics offers a practical pathway to address delayed identification of internalizing symptoms while balancing accuracy, equity, and feasibility. Recommendations include real-world piloting, explainable AI integration, and stakeholder collaboration to strengthen mental health support systems in underserved schools.