설명가능한 인공지능 알고리즘을 이용한 노인의 자살 생각 예측

초록

This study aims to leverage machine learning models within artificial intelligence to predict the characteristics of older adults that contribute to suicidal ideation through big data analysis. Furthermore, it seeks to expand the existing body of research on factors influencing suicidal thoughts. To achieve this objective, the study evaluated the predictive performance of six machine learning and deep learning algorithms. The findings indicate that the LightGBM (LGBM) algorithm demonstrated the highest predictive accuracy, achieving 98.74%, while precision was maximized at 99.79% when combined with the Random Forest algorithm. These results suggest that LGBM can accurately identify suicidal ideation in more than 99 out of 100 older adults. To further interpret the model's predictions, this study employed the SHAP (SHapley Additive exPlanations) model, which offers the advantage of analyzing both individualized and overall contributing factors within machine learning and deep learning frameworks. The analysis identified key variables associated with suicidal ideation in later life, including heightened levels of sadness, exposure to partner violence, current smoking status, sleep disturbances, outpatient visits to medical services within the past year, decreased meaningful discussions within the family, frequent loss of appetite, lower satisfaction with family income, advanced age, and reduced grocery expenditures. By utilizing the complete dataset from the Korean Welfare Panel spanning the past decade, this study applied explainable AI algorithms to specify demographic, economic, social, and health-related factors influencing suicidal ideation. The findings provide practical information for targeted suicide prevention strategies.

키워드