AI-Driven Prediction of Mental Disorders: Enhancing Early Diagnosis and Intervention in the USA

Abstract

Early detection of mental disorders remains one of the most pressing challenges in U.S. public health, as socioeconomic and behavioral indicators often precede clinical diagnosis but are rarely integrated into predictive frameworks. This study develops an AI-driven diagnostic pipeline that fuses demographic, behavioral, and social determinants of health to predict risk for major mental disorders, including anxiety, depression, and post-traumatic stress disorder (PTSD). Using a population-scale dataset of over 10,000 anonymized health records combining age, sex, BMI, income, education, and lifestyle behaviors, we benchmark five machine learning models, Logistic Regression, Random Forest, XGBoost, Support Vector Machine (SVM), and a Multi-Layer Perceptron (MLP), across both unbalanced and SMOTE-balanced conditions. The evaluation integrates multiple dimensions: discrimination (ROC-AUC, PR-AUC), calibration (Brier score, reliability curves), and fairness across demographic subgroups (sex, rural–urban classification). SHAP-based explainability is employed to interpret model behavior and to identify dominant risk predictors and interaction effects, while robustness checks probe performance under covariate shifts and synthetic missingness. Results show that ensemble and deep models outperform classical baselines, with XGBoost achieving an average ROC-AUC of 0.90 and strong calibration stability. Income level, alcohol consumption, and BMI category emerge as top predictors, reflecting known epidemiological associations. Subgroup analysis demonstrates consistent performance across demographic segments, underscoring model fairness and generalizability. Collectively, the findings illustrate how interpretable AI can enhance early detection and risk stratification for mental health conditions, providing a data-driven foundation for preventive interventions, policy guidance, and equitable digital mental health systems in the United States.