Machine learning based suicide prediction and development of suicide vulnerability index for US counties

Area of Science:

• Public Health
• Data Science
• Computational Epidemiology

Background:

• Suicide presents a significant and growing public health challenge in the United States.
• Understanding and predicting suicide patterns are crucial for effective control and prevention strategies.

Purpose of the Study:

• To analyze suicide trends and geographical distribution across US counties from 2010-2019.
• To develop and validate a machine learning model for county-level suicide prediction.
• To identify key demographic features influencing suicide rates.

Main Methods:

• Utilized publicly available data spanning 2010-2019 for all 3140 US counties.
• Developed an eXtreme Gradient Boosting (XGBoost) machine learning model incorporating 17 features.
• Employed SHapley Additive exPlanations (SHAP) to determine feature importance.

Main Results:

• Observed significant increases in suicide rates in numerous counties; approximately 25% saw at least a 10% rise, and 12% experienced over a 50% increase.
• The XGBoost model achieved a high predictive accuracy with an R² value of 0.98.
• Total Population, % African American, % White, Median Age, and % Female Population were identified as the top 5 predictors.

Conclusions:

• A novel Suicide Vulnerability Index (SVI) was developed using the top 5 predictive features.
• The SVI can identify US counties at high risk for suicide.
• This tool supports informed decision-making for targeted suicide prevention and control initiatives.