Analysis and evaluation of explainable artificial intelligence on suicide risk assessment

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Mental Health Research

Background:

• Healthcare Machine Learning (ML) often faces limited datasets.
• Accurate suicide risk prediction is crucial for timely intervention.
• Explainable Artificial Intelligence (XAI) offers insights into complex ML models.

Purpose of the Study:

• To evaluate the effectiveness of XAI in predicting suicide risk from medical data.
• To enhance suicide risk identification using ML and data augmentation.
• To identify key factors influencing suicide risk and potential preventive measures.

Main Methods:

• Utilized Machine Learning (ML) models, including Random Forest (RF).
• Employed data augmentation techniques to address dataset limitations.
• Applied SHapley Additive exPlanations (SHAP) for XAI and correlation analysis for feature importance.

Main Results:

• The RF model achieved high accuracy, F1 score, and AUC (>97%).
• SHAP analysis identified anger issues, depression, and social isolation as primary suicide risk predictors.
• Factors like high income, professional status, and education correlated with lower risk.

Conclusions:

• ML and XAI are effective tools for suicide risk assessment.
• Findings provide valuable insights for psychiatrists and clinical decision-making.
• This approach can aid in developing targeted suicide prevention strategies.