Abstract This study developed a multidimensional machine learning model grounded in ecological systems theory to enable early depression detection and targeted interventions among Chinese adolescents. Utilizing data from 7,169 middle school students (aged 14.98 ± 1.58 years, depression prevalence 36.01%) in Liaoning Province, China, 21 key predictors were selected via LASSO regression and Boruta algorithm. Five models (XGBoost, random forest, logistic regression, multilayer perceptron, and support vector machine) were evaluated. XGBoost demonstrated optimal performance (AUC = 0.912). SHAP analysis identified five core predictors: sleep quality (primary factor), authenticity, meaning in life, sense of control, and perceived peer-related stress. Protective factors included sleep quality, authenticity, and sense of control, while perceived peer-related stress was a risk factor. Nonlinear associations emerged between meaning in life and depression, with age-stratified thresholds (12–13 years: moderate meaning linked to highest risk; 14–16 years: moderate meaning reduced risk; 17–18 years: high meaning mitigated risk). Findings suggest a tri-level intervention framework: sleep-exercise programs (biological), age-specific resilience training (cognitive), and AI-driven school stress monitoring (environmental).
Li et al. (Sun,) studied this question.