To construct a predictive model for non-suicidal self-injury (NSSI) among adolescents using machine learning techniques, addressing a critical public health issue.
Key Findings:
NSSI incidence rates were approximately 24%, 23%, and 22% at T1, T2, and T3, respectively.
The SVM model outperformed others with AUC values greater than 0.75 and recall and F1 scores above 0.7 at all time points.
Key predictors of NSSI risk included suicide-related ideation, school bullying, and depressive status.
Interpretation:
The SVM model effectively predicts NSSI risk in adolescents, highlighting the importance of suicide-related behaviors as significant predictors, which can inform targeted interventions.
Limitations:
The study's retrospective design may limit causal inferences.
Findings may not be generalizable beyond the studied population.
Potential biases inherent in retrospective studies may affect results.
Conclusion:
The study provides a robust predictive framework for identifying adolescents at high risk of NSSI, aiding in the development of targeted prevention strategies.
