Machine learning-based analysis and prediction of factors influencing mental health among children and adolescents in Jiangsu Province.

Background: This study investigates the current mental health status among children and adolescents in Jiangsu Province by analyzing symptoms of depression, anxiety, and stress using standardized psychological scales. Machine learning models were utilized to identify key influencing variables and predict mental health outcomes, aiming to establish a rapid psychological well-being assessment framework for this population.

Objective: A cross-sectional survey was conducted via random cluster sampling across 98 counties (cities/districts) in Jiangsu Province, enrolling 141,725 students (47,502 primary, 47,274 junior high, 11,619 vocational high school students, and 35,330 senior high ). The study focused on prevalent mental health disorders and associated risk factors.

Methods: Depression, anxiety, and stress scores served as dependent variables, with 57 socio-demographic and behavioral factors as independent variables. Five supervised machine learning models (Decision Tree, Naive Bayes, Random Forest, K-Nearest Neighbors (KNN), and XGBoost) were implemented using R software. Model performance was evaluated using accuracy, precision, recall, F1 Score and Area Under the ROC Curve (AUC). Feature importance analysis was conducted to identify key predictors.

Results: The study revealed significant mental health disparities: depression (14.9%), anxiety (25.5%), and stress (10.9%) prevalences showed clear gender and regional gradients. Females exhibited higher rates across all conditions (p < 0.05), and urban areas had elevated risks compared to suburban regions. Mental health deterioration escalated with educational stages (e.g., depression from 9.2% in primary to 21.2% in senior high; χ²_trend = 2274.55, p < 0.05). The XGBoost model demonstrated optimal predictive performance (AUC: depression = 0.799, anxiety = 0.770, stress = 0.762), outperforming other models. Feature importance analysis consistently identified bullying duration, age, and drinking history as top risk factors across both Gain and SHAP methods, while SHAP values additionally emphasized modifiable lifestyle factors (e.g., breakfast frequency) and demographic variables (e.g., gender).

Conclusions: This study identifies bullying, age, and alcohol consumption history as key mental health risk factors among Jiangsu's children and adolescents. These findings emphasize the need for school-based anti-bullying programs, age-specific mental health counseling, and healthy lifestyle education (including alcohol refusal). Lifestyle behaviors like daily breakfast intake should be integrated into dietary interventions for mental health promotion. Urban-rural and gender disparities necessitate targeted support for urban adolescent females, while educational stage differences highlight the criticality of early prevention.