Machine learning-based classification and risk factor analysis of frailty in Korean community-dwelling older adults

Frailty is a dynamic reversible state, characterized by frequent transitions between frailty status over time. The timely and effective detection of frailty is important to prevent adverse health outcomes. This study aims to develop machine learning-based classification models for frailty assessment and to investigate its risk factors. A total of 1,482 subjects, 1,266 robust and 216 frail older adults, were analyzed. Sixteen frail risk factors were selected from a random forest-based feature selection method, then used for the inputs of five ML models: logistic regression, K-nearest neighbor, support vector machine, gaussian naïve bayes, and random forest. Data resampling, stratified 10-fold cross-validation, and grid search were applied to improve the classification performance. The logistic regression model using the selected features showed the best performance with an accuracy of 0.93 and an F 1 -score of 0.92. The results suggest that machine learning techniques are an effective method for classifying frailty status and exploring frailty-related factors.Clinical Relevance— Our approach can predict frailty using data collectable in clinical setting and can help prevent and improve by identifying variables that change frailty status.