Abstract Medical datasets often contain numerous redundant or noisy features, which can degrade classification performance and increase computational costs. Feature selection (FS) is therefore crucial for improving diagnostic accuracy and enhancing model interpretability in biomedical applications. Although the recently introduced enzyme action optimizer (EAO) has shown strong potential as a metaheuristic method, its effectiveness for FS and its behavior in high-dimensional spaces remain underexplored. Like many swarm intelligence (SI) algorithms, EAO faces challenges in maintaining population diversity, achieving a balanced exploration–exploitation process and avoiding premature convergence. To address these limitations, this study proposes an enhanced enzyme action optimizer (EEAO) for FS. The method integrates latin hypercube sampling to ensure diverse initialization, refraction learning to enhance global exploration and a crowding-distance mechanism to reduce solution clustering and improve search stability. A binary version of EEAO is further developed using an S-shaped transfer function to efficiently handle FS tasks. Due to the iterative nature of the algorithm and the high dimensionality of biomedical datasets, EEAO-FS is computationally intensive; implementing it using parallel, distributed or HPC-enabled frameworks can significantly accelerate convergence and enable scalable analysis of large datasets. The method is evaluated against eight state-of-the-art FS methods on twenty biomedical datasets using five performance metrics. Experimental results show that EEAO achieves the highest average accuracy (91.85%), selects substantially fewer features (approximately 60% reduction), converges more consistently and requires less execution time compared to competing methods. Statistical analysis confirms the significance of these improvements. EEAO’s global optimization performance is validated on eight benchmark functions, showing superior exploration, faster convergence and robustness. These results demonstrate its efficiency and reliability for high-dimensional medical FS while highlighting its suitability for high-performance and scalable computational frameworks.
Hegazy et al. (Thu,) studied this question.