Ensemble Feature Selection for Elderly Health Condition Classification

Abstract

Feature selection is a pivotal technique for constructing classification models with high-dimensional data. It facilitates the identification and elimination of redundant or irrelevant features, thereby reducing model complexity and enhancing predictive accuracy. The success of a classification model largely depends on the quality of the selected features. Accordingly, feature selection is performed during data preparation, prior to data processing, to support the classification of large-scale datasets. This study proposes an ensemble feature selection strategy based on embedded methods to identify the most informative feature subset for classifying elderly health conditions. The selected features include clinical indicators from Activities of Daily Living (ADL) and Comprehensive Geriatric Assessment (CGA), enabling a holistic evaluation of functional status. The proposed approach achieves an impressive classification accuracy of 98.11%. The resulting optimal feature subset provides a valuable foundation for planning and advancing elderly healthcare and rehabilitation.