Predicting the Livability of Older Adults Using a Livability Scale Through Machine Learning: A Pilot Study

The interaction of multidimensional person, environment, task, and performance factors requires advanced analytic approaches to understand livability among older adults. This study examined the predictive value of the Livability Scale (LS) for residential satisfaction using multiple machine learning algorithms in adults aged 65 and above. Six models-logistic regression, decision tree, random forest, gradient boosting, support vector machine, and an ensemble model-were compared to determine which best captured the LS complex structure. The random forest demonstrated the strongest performance (F1 = 0.74, AUC = 0.78) with consistent generalization in k-fold cross-validation (M = 0.68), providing balanced sensitivity (0.60) and specificity (0.90). Model comparison also revealed meaningful differences: SVM showed heightened sensitivity to satisfaction, correctly detecting a larger number of satisfied cases but with increased false positives, whereas GBM minimized false positives, indicating higher specificity but lower sensitivity. Feature-importance analysis identified cultural facilities, residential affordability, activities and events, parks, structural safety, and cleanliness as key contributors to predictions. Overall, the findings emphasize the value of machine learning in uncovering non-linear patterns in livability data and demonstrate the potential of algorithmic diversity to guide evidence-based residential interventions for older adults.