Comparison of Machine Learning Models for Predicting Suicide Attempts among Korean Adolescents with Suicidal Ideation: Secondary Data Analysis Based on 20th Korea Youth Risk Behavior Survey

Purpose: This study aimed to identify the factors associated with the progression from suicidal ideation to suicide attempts and to compare the predictive performance of various machine learning models. Methods: We conducted a secondary analysis using original data from the 20th Korea Youth Risk Behavior Survey (KYRBS), focusing on 6,316 adolescents who reported suicidal ideation. We evaluated predictive performance using logistic regression, random forest, and k-nearest neighbors (KNN) models. Results: Suicide attempts were significantly associated with sociodemographic factors, such as academic achievement, economic status, type of residence, and perceived health status, as well as psychological and behavioral factors, including suicidal planning, feelings of sadness and despair, anxiety, alcohol use, smoking, drug use, and exposure to violence. Logistic regression exhibited the highest predictive performance (AUC=0.77, accuracy=0.84, F1=0.80). The random forest model identified suicidal planning, loneliness, generalized anxiety, drug use, and exposure to violence as key predictors based on the Gini index, while KNN demonstrated the lowest predictive stability. Conclusion: Logistic regression is effective for predicting suicide attempts among adolescents, and machine learning approaches should be considered for early risk screening in community mental health nursing.