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

By
Yujun Zhao
Qian Wang
Wei Liu
May 13, 2026
0 min

Frontiers In Psychiatry

Overview

This study developed a machine learning model to predict non-suicidal self-injury (NSSI) in adolescents, identifying key risk factors such as suicide-related ideation and bullying. The support vector machine model outperformed others in predictive accuracy, highlighting the importance of early identification and intervention.

Background

Non-suicidal self-injury (NSSI) is a prevalent issue among adolescents, with significant implications for mental health and suicide risk. Understanding and predicting NSSI can facilitate timely interventions, potentially reducing the incidence of more severe outcomes such as suicide. This study leverages machine learning to enhance predictive capabilities beyond traditional methods.

Data Highlights

Time Point	Incidence Rate of NSSI
T1	24%
T2	23%
T3	22%

Key Findings

The SVM model showed superior performance with AUC values greater than 0.75.
Recall and F1 scores for the SVM model were both higher than 0.7.
Key predictors of NSSI included suicide-related ideation, school bullying, and depressive status.
The incidence rates of NSSI were approximately 24%, 23%, and 22% at T1, T2, and T3, respectively.
SHAP analysis confirmed the importance of psychological and environmental factors in predicting NSSI risk.

Clinical Implications

The findings underscore the necessity for clinicians to assess risk factors such as bullying and depressive symptoms in adolescents. Implementing machine learning models in clinical settings may enhance the identification of at-risk youth, allowing for targeted prevention strategies.

Conclusion

The study demonstrates the effectiveness of machine learning in predicting NSSI among adolescents, emphasizing the role of psychological factors in risk assessment. Early identification can lead to improved intervention strategies.