Prediction of Recurrent Falls in Community-dwelling Older Adults Using Machine Learning Techniques

Objectives: This study aimed to develop a machine learning-based model to predict recurrent falls in community-dwelling older adults aged 65 and over who have a history of falling and to identify key predictors to inform early detection and preventive strategies. Methods: Data from 612 older adults with prior falls, drawn from the 2023 Korean National Survey of Older Persons, were analyzed. Recurrent falls were defined as experiencing two or more falls in the past year. Independent variables included sociodemographic, health status, health behavior, functional status, fall-related, and environmental factors. Four models—Logistic regression, Random forest, XGBoost, and CatBoost—were used. Model performance was assessed using accuracy, precision, recall, F1-score, and area under the curve (AUC). Feature importance was analyzed based on the best-performing model. Results: Random forest demonstrated the overall best performance and showed a significant difference compared with logistic regression, indicating that it was the most suitable model for predicting recurrent falls in this study. The major predictors of recurrent falls were identified as age, depression, fall treatment, cognitive function, and fall reason. Conclusions: This study is the first to apply machine learning for recurrent fall prediction among older adults in Korea, presenting models that demonstrated superior performance compared to traditional regression analysis. The findings provide an academic foundation for recurrent fall prediction research in Korea and confirm the feasibility of early identification of high-risk groups. Future work should focus on advancing the model to develop a decision-support tool applicable in both community and clinical settings.