Comparative Analysis of Machine Learning Algorithms for Predicting Health and Living Conditions of Elderly Living Alone: Focusing on the 2023 National Survey of Older Koreans
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
(Purpose) This study systematically compared machine learning algorithms for predicting health and living conditions of elderly living alone, using data from the 2023 National Survey of Older Koreans. The research developed a multi-outcome model to simultaneously predict depression risk, subjective health, and economic difficulty, and explored differences by gender, age, region, and education. (Design/methodology/approach) Five classification models—logistic regression, support vector machine (SVM), decision tree, artificial neural network, and AdaBoost—were applied to 2,487 elderly living alone from the 2023 survey. Model performance was evaluated by accuracy, precision, recall, F1-score, and AUC. Stability was checked using 5-fold cross-validation, with key risk factors identified via SHAP analysis. (Findings) SVM demonstrated the strongest predictive accuracy and F1-score for depression risk among the elderly living alone, while AdaBoost delivered nearly equivalent performance with notably reduced computational time; subgroup analyses further clarified that risk factors such as social isolation, economic stress, and daily functioning varied significantly according to gender, age group, and region. (Research implications or Originality) The study underscores the utility of advanced machine learning for detecting complex, multidimensional risk patterns and subgroup heterogeneity among older adults, highlighting the potential for more precise, evidence-based policy and customized welfare service planning aimed at the unique needs of high-risk elderly populations.
Abstract
(Purpose) This study systematically compared machine learning algorithms for predicting health and living conditions of elderly living alone, using data from the 2023 National Survey of Older Koreans. The research developed a multi-outcome model to simultaneously predict depression risk, subjective health, and economic difficulty, and explored differences by gender, age, region, and education. (Design/methodology/approach) Five classification models—logistic regression, support vector machine (SVM), decision tree, artificial neural network, and AdaBoost—were applied to 2,487 elderly living alone from the 2023 survey. Model performance was evaluated by accuracy, precision, recall, F1-score, and AUC. Stability was checked using 5-fold cross-validation, with key risk factors identified via SHAP analysis. (Findings) SVM demonstrated the strongest predictive accuracy and F1-score for depression risk among the elderly living alone, while AdaBoost delivered nearly equivalent performance with notably reduced computational time; subgroup analyses further clarified that risk factors such as social isolation, economic stress, and daily functioning varied significantly according to gender, age group, and region. (Research implications or Originality) The study underscores the utility of advanced machine learning for detecting complex, multidimensional risk patterns and subgroup heterogeneity among older adults, highlighting the potential for more precise, evidence-based policy and customized welfare service planning aimed at the unique needs of high-risk elderly populations.