Benchmarking Binary Metaheuristics for Joint Feature Selection and SVM Tuning in Biomedical Classification

Hyperparameter optimization and feature selection are closely related aspects of biomedical classification systems, especially when samples are high-dimensional and limited in sample size. Traditional pipelines normally tackle these components separately, though they interact heavily in kernel-based classifiers like Support Vector Machines (SVMs). This paper presents a unified optimization model that concurrently selects informative biomedical features and optimizes SVM hyperparameters, and compares three high-level binary metaheuristic algorithms (Binary Harris Hawks Optimization (BHHO), Binary Differential Evolution (BDE), and Binary Marine Predators Algorithm (BMPA)) in the same set of experimental conditions. Both candidate solutions combine a binary feature-selection vector with continuous SVM parameters to allow exploration of the mixed search space. The proposed framework is tested against various biomedical data to perform different diagnostic tasks. The evaluation is performed experimentally in terms of AUC, F1-score, accuracy, and the feature reduction rate. Results demonstrate consistent improvements over baseline methods, with the highest overall discriminative performance recorded with BHHO, while BMPA produces the most compact feature subsets. This result shows that combined optimization using the proposed binary metaheuristics provides a powerful and interpretable method of diagnosing biomedical problems.