The concise machine learning prediction models for suicide attempt in China: Based on demographic and social factors.

Area of Science:

• Computational psychiatry
• Clinical informatics
• Public health research

Background:

• Machine learning (ML) methods are increasingly used for suicide attempt (SA) prediction.
• Limited research exists on ML models for Chinese populations, often using overly complex variables.
• The need for concise and applicable predictive models is critical.

Purpose of the Study:

• To explore the efficacy of ML approaches in predicting suicide attempts.
• To develop a more concise and applicable predictive model for SA.
• To compare the performance of multiple ML algorithms in a Chinese cohort.

Main Methods:

• A case-control survey in China collected demographic data and utilized the GSS Suicide Attitude Scale and Beck Scale for Suicide Ideation.
• Five ML methods—Random Forest (RF), MLR, XGBoost, AdaBoost, and LightGBM—were employed.
• Model performance was evaluated using standard indices, with R 4.2.1 software facilitating the analysis.

Main Results:

• Multiple ML models achieved Area Under the Curve (AUC) values greater than 0.75, indicating good predictive efficiency.
• The RF model achieved the highest overall AUC (0.8638) and optimal accuracy (79.09%) on the test dataset.
• LightGBM showed the highest AUC in the training dataset (0.9199) and highest positive predictive value.

Conclusions:

• All evaluated ML algorithms performed well in distinguishing between SA cases and non-SA cases.
• RF and LightGBM emerged as the top-performing models in this study.
• Future research should explore ensemble or combined model strategies to further enhance prediction accuracy and detection rates.