Using machine learning approach to predict suicide ideation and suicide attempts among Chinese adolescents: A cross-sectional study

Background: Screening for suicide ideation and suicide attempts is crucial for adolescents, yet accurately predicting these outcomes remains a significant challenge. The relationship between non-suicidal self-injury and suicide ideation and attempts is complex. Therefore, this study employs machine learning techniques to explore and address these challenges. Methods: Using cross-sectional data, this study used stratified cluster sampling to recruit 2123 adolescents aged 10 to 15 from Fuyang, Anhui Province. Five psychological scale scores and six sociodemographic parameters were collected and included in eight machine learning models for factor selection and prediction of suicide ideation and suicide attempts. Results: The XGBoost model achieved an impressive AUC of 0.89 for suicide ideation and 0.92 for suicide attempts. Feature importance analysis revealed that suicide attempts are the most critical feature influencing suicide ideation, with an importance score of 0.73. Conversely, suicide ideation was identified as a significant predictor of suicide attempts, with an importance score of 0.86. Furthermore, non-suicidal self-injury was also found to significantly impact both outcomes. Notably, the relationship between the frequency of non-suicidal self-injury and both suicide ideation and attempts are nonlinear. Limitation: A key limitation is that this cross-sectional study somewhat restricts the ability to establish causal relationships. Conclusion: This study applied machine learning techniques to identify critical factors influencing adolescent suicide ideation and attempts, uncover the intricate relationship between non-suicidal self-injury and these outcomes, enhance risk prediction accuracy, and support early screening and intervention efforts.