Comprehensive Summary
This study used machine learning to predict non-suicidal self-injury (NSSI) in 3,483 Chinese adolescents over a 2.5-year period. Researchers collected psychological, behavioral, social, and family data across four time points and compared seven machine-learning models. The Random Forest model showed the strongest and most consistent accuracy in predicting which students were at greatest risk. Using SHapley Additive exPlanation (SHAP) analysis, the study identified major risk factors—such as suicide-related behaviors, depression, anxiety, delinquent behaviors, and family dysfunction—and key protective factors like spirituality, emotional competence, empathy, and life satisfaction. By combining longitudinal data with explainable machine learning, the study demonstrated that NSSI risk can be accurately predicted over both short- and long-term periods.
Outcomes and Implications
This research shows how machine-learning tools could transform mental-health care by allowing earlier and more accurate identification of youth at risk of self-injury. Instead of relying on clinical judgment alone, schools and healthcare providers could use predictive models to screen large populations and intervene before self-harm occurs. Because the model is explainable, clinicians can also see which factors, such as depression, anxiety, or lack of emotional coping skills, are driving a patient’s risk, allowing for personalized treatment and targeted prevention programs. In the long term, integrating predictive analytics into healthcare could support better mental health monitoring, reduce emergency hospitalizations, and inform public health policy aimed at protecting vulnerable adolescents.