Predicting Care Needs in Community-Dwelling Older Adults Using Explainable Machine Learning and a Multidimensional Approach Integrating Health, Social, and Environmental Factors: Cross-Sectional Study.

Background: Rapid population aging and a worsening shortage of care workers necessitate the identification of older adults who require proactive interventions. Although machine learning (ML) has been increasingly applied in gerontology, existing studies have predominantly focused on social isolation, loneliness, depression, falls, and frailty in isolation rather than on the integrated construct of care needs. Objective: This study aimed to develop and interpret an explainable ML model that identifies care needs in community-dwelling Korean older adults. Beyond physical health indicators such as disease and functional status, this study adopted a comprehensive approach that included mental health, cognitive function, health behaviors, and socioenvironmental determinants, such as social participation, social support, and the housing environment, to present an integrated model encompassing both health and social care needs. Methods: Data were obtained from the 2023 Korea Senior Survey, a nationally representative sample of 10,078 community-dwelling adults aged 60 years and older. The data were split 70:30 into training (n=7054) and held-out test (n=3024) sets. Seven algorithms were compared (logistic regression, decision tree, support vector machine, random forest, gradient-boosted decision trees, extreme gradient boosting, and light gradient boosting machine) using stratified 5-fold cross-validation on the training set. Discrimination was assessed by the area under the receiver operating characteristic curve (AUC). Model interpretability used Shapley additive explanations with bootstrap stability assessment across folds. Results: Model A (excluding activities of daily living or instrumental activities of daily living [IADL]) achieved good discrimination (AUC 0.892, 95% CI 0.866-0.918), adequate calibration (calibration slope=0.826), and positive clinical net benefit, demonstrating that upstream factors alone can identify older adults with care needs without relying on functional status. Shapley additive explanations analysis identified age, nutritional risk, employment status, depressive symptoms, self-rated health, cognitive function, household income, and home modification as the leading predictors, with high rank stability across cross-validation folds. Model B (including activities of daily living or IADL) yielded a higher AUC (0.976, 95% CI 0.962-0.989), but this reflected the near-tautological relationship between IADL and self-reported care needs rather than genuine upstream predictive value. Using an objective composite outcome yielded equivalent discrimination (AUC 0.892), supporting robustness to the outcome definition. Conclusions: Explainable ML models offer high predictive accuracy and strong interpretability for identifying care needs among older adults. Care needs in community-dwelling Korean older adults can be identified with good discrimination, calibration, and clinical net benefit using multidimensional nonfunctional factors alone. By highlighting the significant roles of health, social, and environmental factors, this study provides empirical evidence to support evidence-based decision-making for the Long-Term Care Insurance system and integrated community care policies.