Objective
This pilot study aimed to assess the feasibility of building predictive models for suicidal ideation in adolescents by integrating key factors from the integrated motivational-volitional model of suicidal behaviour (IMV).
Methods
Based on IMV model and prior research, 37 factors were selected for a cluster sampling survey. A total of 1,517 Chinese students in grades 7-9 completed self-report questionnaires. We employed logistic regression, XGBoost, decision tree, and random forest algorithms to develop models predicting active suicidal ideation. Synthetic minority over-sampling technique and 10-fold cross-validation were used to deal with imbalanced data and prevent over-fitting. We used receiver operating characteristic (ROC) to evaluate the reliability of models.
Results & Discussion
The ROC analysis revealed that logistic regression yielded a larger AUC for predicting suicidal ideation compared to XGBoost, decision tree, and random forest (0.900 vs. 0.880, 0.790 and 0.890, respectively). Across all models, "perceived burdensomeness" and "entrapment" were consistently identified as two of the three most influential variables. Additionally, life events, depressive symptoms, gender (female), internet addiction, hopelessness, anxiety, and defeat were recognized as significant factors in multiple models.
Conclusions
This pilot study demonstrated that the IMV model contribute to the construction of effective predictive models of suicidal behaviour. However, the limited sample size data and the absence from other domains limited the potential of machine learning algorithms to outperform regular logistical regression.
Acknowledgement: This work was supported by the the Fundamental Research Funds for the Central Universities (2412024QD035).
 

adolescent suicidal ideation,machine learning,Based on IMV model