Development of Machine Learning Models to Categorize Life Satisfaction in Older Adults in Korea

Objectives: This study aimed to identify factors associated with life satisfaction by developing machine learning (ML) models to predict life satisfaction in older adults living alone. Methods: Data were extracted from 3112 older adults participating in the 2020 Korea Senior Survey. We employed 5 ML models to classify the life satisfaction of older adults living alone: logistic Lasso regression, decision tree-based classification and regression tree (CART), C5.0, random forest, and extreme gradient boost (XGBoost). The variables used as predictors included demographics, health status, functional abilities, environmental factors, and activity participation. The performance of these ML models was evaluated based on accuracy, precision, recall, F1-score, and area under the curve (AUC). Additionally, we assessed the significance of variable importance as indicated by the final classification models. Results: Out of the 1411 older adults living alone, 45.3% expressed satisfaction with their lives. The XGBoost model surpassed the performance of other models, achieving an F1-score of 0.72 and an AUC of 0.75. According to the XGBoost model, the five most important variables influencing life satisfaction were overall community satisfaction, self-rated health, opportunities to interact with neighbors, proximity to a child, and satisfaction with residence. Conclusions: Overall satisfaction with the community environment emerged as the most significant predictor of life satisfaction among older adults living alone. These findings indicate that enhancing the supportiveness of the community environment could improve life satisfaction for this demographic.