A novel swarm intelligence-based approach for feature selection in sarcopenia precision modeling

Sarcopenia is a multifactorial skeletal muscle disorder associated with aging and is strongly linked to frailty, functional decline, disability, and increased healthcare burden among older adults. However, accurate early prediction remains challenging due to the high dimensionality, heterogeneity, and complex interactions among clinical variables. This study aimed to develop a swarm intelligence-based feature selection framework to improve predictive performance and enhance interpretability for precision sarcopenia risk prediction. A novel feature selection algorithm, termed the Flying-Geese Migration Optimization Algorithm (FGMOA), was proposed to identify informative and compact subsets of clinical features. Clinical data were obtained from the sarcopenia assessment database at Kaohsiung Medical University and underwent standardized preprocessing prior to model development. The optimization capability of FGMOA was evaluated against five established swarm intelligence algorithms, including Grey Wolf Optimizer (GWO), Whale Optimization Algorithm (WOA), Salp Swarm Algorithm (SSA), Fox Optimization Algorithm (FOA), and Honey Badger Algorithm (HBA). The selected feature subsets were subsequently used to develop predictive models based on three ensemble learning classifiers: Gradient Boosting Decision Trees (GBDT), Extreme Gradient Boosting (XGBoost), and CatBoost. Model performance was evaluated using accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic curve (AUC). The proposed FGMOA framework demonstrated superior predictive performance compared with the evaluated swarm intelligence algorithms across different classifier settings. Among the developed models, FGMOA-XGBoost achieved the best overall performance, with an accuracy of 96.78%, a recall of 97.32%, a precision of 96.04%, and an F1-score of 96.67%. Furthermore, FGMOA-CatBoost achieved the highest recall, 98.21%, indicating strong sensitivity in identifying individuals at risk of sarcopenia. These findings indicate that FGMOA can effectively identify informative feature subsets and improve classification performance in high-dimensional clinical data. The proposed FGMOA-based framework provides a promising approach for feature selection and sarcopenia risk prediction by integrating swarm intelligence optimization with ensemble learning. The framework effectively reduces feature dimensionality while maintaining favorable predictive performance and interpretability. Although further validation across independent, multicenter datasets is required, this approach may contribute to the development of data-driven tools for precision geriatric healthcare and early assessment of sarcopenia risk.