Optimizing Suicide Risk Prediction in Korea: A Comparison of Model Performance Using Resampling Methods and Machine...

Area of Science:

• Public Health
• Data Science
• Psychiatry

Background:

• Suicide risk prediction is crucial for public health interventions.
• Machine learning (ML) offers potential for developing accurate suicide risk prediction models.
• Optimizing ML models requires careful selection of algorithms and data preprocessing techniques.

Purpose of the Study:

• To develop an optimal Korean suicide prediction model using various ML algorithms and resampling methods.
• To evaluate the performance of different ML models in predicting suicidal ideation.
• To identify effective strategies for handling data imbalance in suicide risk prediction.

Main Methods:

• Utilized data from the Korea National Health and Nutrition Examination Survey (2017, 2019, 2021) for individuals aged 19 years and older.
• Applied five ML algorithms: logistic regression, random forest (RF), k-nearest neighbor, gradient boosting, and adaptive boosting.
• Implemented undersampling and oversampling techniques to address data imbalance issues.

Main Results:

• The random forest (RF) model trained with undersampled data demonstrated superior performance.
• Achieved a sensitivity of 0.781 and an area under the curve (AUC) of 0.870 with the optimal RF model.
• Identified the RF model as highly effective for predicting suicidal ideation in the Korean population.

Conclusions:

• The developed ML model provides a promising tool for predicting suicide risk in Korea.
• Further validation of this ML model can enhance the prediction of suicide risk factors.
• Integrating individual, social, and environmental factors into ML models can improve suicide risk assessment.