Machine-Learning Prediction of Health-Related Quality of Life Among Community-Dwelling Middle-Aged and Older Adults Living Alone: A Secondary Analysis of the 2022 Korea Health Panel

Background/Objectives: Because the numbers of middle-aged and older adults living alone in Korea have substantially increased, which warrants greater attention to their health-related quality of life. Therefore, we aimed to develop a predictive model for the health-related quality of life among community-dwelling middle-aged and older adults living alone. Methods: Using 2022 Korea Health Panel Survey data, 1313 participants with complete EQ-5D component data were analyzed. All candidate predictors were entered into benchmarked models without pre-model feature selection. Preprocessing and 5-fold cross-validated hyperparameter tuning were conducted within the training data. Final performance was evaluated on a held-out test set, and the selected model was interpreted using SHAP. Results: XGBoost had the lowest training cross-validated RMSE and was selected as the final explainable model. On the test set, it showed moderate performance (R2 = 0.373, MAE = 0.070, RMSE = 0.096), outperforming the mean baseline model (RMSE = 0.121) but remaining comparable with other top-performing models. Predictions were within absolute errors of 0.05 and 0.10 for 45.6% and 76.4% of participants, respectively. SHAP ranked subjective health, age, walking time, need for care, and monthly household income as the five highest-ranked predictors. Other highly ranked predictors included unmet medical needs, total annual out-of-pocket expenditure, disability, anxiety, and regular exercise. Conclusions: These findings may inform targeted interventions and support strategies, although external validation and longitudinal studies are needed to confirm generalizability and causal relationships.