Abstract

We developed and deployed a prototype, an effective prediction model to identify college students at risk of Non-Suicidal Self-Injury (NSSI), utilizing various supervised machine-learning algorithms. The sample was obtained from the records of 4,956 college students who used counseling services. After addressing missing data and balancing the data with the SMOTE technique, the final sample consisted of 1,484 students, of whom 742 exhibited NSSI (positive class) and 742 did not have a history of NSSI (negative class). Statistical and clinical importance analyses were used to choose seven predictor factors. Four models were trained and evaluated using supervised machine-learning techniques, and hyperparameters were adjusted. The best model (Model 1 with SVM) had a sensitivity of 0.83 and a precision of 0.71. This indicates a substantial ability to detect students with NSSI, with a good balance between true and false positives, according to the F1 score of 0.77. These findings suggest that implementing these predictive tools in college counseling services could greatly improve the detection and prevention of NSSI. The potential of machine learning to enhance early intervention in mental health issues is emphasized, urging future research to expand its applicability and effectiveness in various educational settings. Additionally, we introduced a Streamlit-based prototype interface that enables counselors to input key predictors, generate NSSI risk predictions, and receive probability metrics in real time, demonstrating practical feasibility for streamlined early intervention.