Identifying key features of suicide attempts in young adults with nonsuicidal self-injury: a machine learning approach

Individuals who engage in nonsuicidal self-injury (NSSI) are at significantly elevated risk for suicide attempts (SA). However, distinguishing which individuals within this high-risk population have already transitioned to suicidal behavior remains a clinical challenge due to the complex interaction of risk and protective factors. This study applied and compared three machine learning (ML) algorithms to identify key classifying features of SA among young adults with NSSI and to elucidate the relative importance of these features. A total of 601 young adults (aged 19–29 years) who reported a lifetime history of five or more NSSI episodes were recruited via an online survey. We utilized a comprehensive array of variables, including sociodemographic, psychosocial, psychiatric, and psychological features (including emotion regulation, distress tolerance, and resilience), as well as NSSI characteristics (frequency and versatility), to classify SA history. Three ML models (support vector machine [SVM], random forest, and eXtreme Gradient Boosting) were employed, and Shapley Additive Explanations values were used to interpret feature importance. The baseline SVM model demonstrated superior classification performance relative to random forest and eXtreme Gradient Boosting (mean area under the receiver operating characteristic curve [AUROC] = 0.811). Following recursive feature elimination, the optimized SVM model achieved a robust AUROC of 0.865 on the held-out test set. Feature importance analysis identified NSSI frequency as the most salient indicator of SA. Furthermore, the presence of mood disorder diagnoses and resilience were the top features. By leveraging ML algorithms, this study effectively differentiated SA history within a high-risk NSSI population. The findings emphasize the critical need for preemptive clinical interventions targeting NSSI frequency to disrupt the acquired capability for suicide. Furthermore, comprehensive prevention strategies must simultaneously mitigate active mood pathology and build resilience to halt the transition to suicidal behavior.