This study employs machine learning–based explainable artificial intelligence to predict suicidal ideation among older adults and to refine understanding of its associated risk factors. Using a decade of data from the Korean Welfare Panel, we compared the predictive performance of six machine learning and deep learning algorithms. LightGBM (LGBM) achieved the highest overall accuracy (98.74%), while precision reached 99.79% when combined with a Random Forest–based ensemble, indicating that suicidal ideation can be identified in nearly all high-risk older adults. To interpret the models, we applied SHAP (SHapley Additive exPlanations), which allows simultaneous examination of global and individual-level feature contributions. Key factors associated with elevated suicidal ideation included high levels of sadness, exposure to partner violence, current smoking, sleep disturbance, recent outpatient service use, reduced meaningful family communication, frequent loss of appetite, low satisfaction with family income, advanced age, and reduced grocery expenditures. By integrating explainable AI with large-scale longitudinal survey data, this study delineates demographic, economic, social, and health-related determinants of suicidal ideation in later life. The findings provide empirical evidence to inform targeted screening, risk stratification, and the design of precision suicide-prevention strategies for older adults in community and clinical settings.

Authors:
Soomin Shin, Korea National University of Transportation, South Korea

About the Presenter(s)
Dr. Soomin Shin is currently an Assistant Professor in the Department of Social Welfare at Korea National University of Transportation, Chungju, South Korea.

Connect on Linkedin
https://www.linkedin.com/in/soominshin

See this presentation on the full schedule – Tuesday Schedule

Bookmark