Prediction of Depressive Experience Among Korean Adolescents: A Comparative Study of Machine Learning Models

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Objectives: This study aimed to develop and compare machine learning models for predicting depressive experience among Korean adolescents and to identify key predictors associated with depressive experience.Methods: Data from 54,152 adolescents who participated in the 2025 Korea Youth Risk Behavior Web-based Survey were analyzed. Five machine learning algorithms—logistic regression, random forest, XGBoost, LightGBM, and CatBoost—were developed and compared. Model performance was evaluated using discrimination and classification metrics, and variable importance was examined using SHAP (SHapley Additive Explanations).Results: The five machine learning models achieved broadly comparable predictive performance, with AUROC values ranging narrowly from 0.753 to 0.761. Among them, XGBoost achieved the highest AUROC of 0.761 (95% confidence interval=0.753–0.770), though the margin over the other models was small. Perceived stress, subjective health status, ever drinking, sleep satisfaction, weight control efforts, grade, and gender emerged as the key predictors.Conclusions: We developed and validated machine learning–based prediction models for identifying adolescents with depressive experience. XGBoost showed the best predictive performance overall, but the gap with the other algorithms was modest. Given these results, the proposed models could support efforts to identify at-risk adolescents early and may inform the design of future mental health screening and prevention strategies.