Development and validation of a machine learning–based risk prediction model for non-suicidal self-injury in adolescents

Background Non-suicidal self-injury (NSSI) among adolescents has long been an important social issue. This study aims to construct a predictive model for NSSI based on machine learning models. Methods A retrospective cohort study design was adopted, including 588 adolescent patients who received psychological and psychiatric assessments. The occurrence of NSSI behavior was used as the outcome variable. Candidate predictors including demographic characteristics, psychological and emotional status, behavioral characteristics, and peer support were collected. The dataset was randomly divided into a training set and a test set at a ratio of 7:3. By comparing the performance of four machine learning models—multiple logistic regression (MLR), random forest (RF), support vector machine (SVM), and eXtreme Gradient Boosting (XGBoost)—at different time points (T1, T2, T3) using area under the curve (AUC), accuracy, precision, recall, and F1 score, the optimal model was selected. The Shapley additive explanations (SHAP) method was further used to conduct interpretability analysis for the optimal model. Results The incidence rates of NSSI at T1, T2, and T3 were approximately 24%, 23%, and 22%, respectively. The SVM model demonstrated superior discrimination ability and stability in predicting the risk of NSSI among adolescents, with AUC values all greater than 0.75 and recall and F1 scores both higher than 0.7. SHAP analyses at all three time points consistently showed that suicide-related ideation and behaviors, school bullying, and depressive status had high contributions to the prediction of NSSI risk. Conclusion The support vector machine model performed best in predicting NSSI among adolescents. Suicide-related behaviors are important predictors of NSSI. The findings of this study help improve the early identification of adolescents at high risk of NSSI and provide evidence for developing targeted prevention and intervention strategies.