Suicide risk prediction for Korean adolescents based on machine learning

Area of Science:

• Psychiatry and Computational Science
• Adolescent Mental Health Research

Background:

• Traditional risk assessment tools for suicidal behavior are insufficient.
• Machine learning (ML) is increasingly used in psychiatric care for risk prediction.

Purpose of the Study:

• To evaluate ML models for predicting adolescent suicidal behavior using national survey data.
• To compare the performance of six different ML models.

Main Methods:

• Six ML models (LR, DT, SVM, GBM, ET, DRF) were compared.
• SHapley Additive exPlanations (SHAP) and Permutation Feature Importance (PFI) were used for interpretability.
• Interaction analysis examined variable relationships.

Main Results:

• The Gradient Boosting Machine (GBM) model showed the highest predictive accuracy (88%).
• Key predictors identified were stress and depression.
• Lower anxiety correlated with reduced suicide risk at higher depression levels.

Conclusions:

• Integrating ML with national survey data improves adolescent suicide risk prediction.
• Findings support early intervention strategies for at-risk youth.