Dynamic binary swordfish movement optimization algorithm for feature selection

Feature selection (FS) is a critical preprocessing stage in machine learning, particularly when dealing with high-dimensional datasets that often contain redundant or irrelevant features. Inefficient handling of such data increases computational complexity and reduces classifier generalization. Existing binary metaheuristics frequently rely on static control parameters, which can lead to premature convergence and poor adaptability. To address these limitations, this paper proposes the Dynamic Binary Swordfish Movement Optimization Algorithm (DBSMOA), a novel nature-inspired binary optimizer based on the foraging dynamics of swordfish. DBSMOA extends the original Swordfish Movement Optimization Algorithm (SMOA) by incorporating dynamic behavioral adaptation mechanisms that enable a self-adjusting balance between exploration and exploitation throughout the optimization process. The algorithm employs a sigmoid-based probabilistic binary mapping and dynamically reassigns agent roles through a performance-driven elitist strategy to maintain diversity and stability during search. Comprehensive evaluations on 52 benchmark datasets show that DBSMOA achieves the lowest average classification error on 35 datasets and the smallest average subset size on 34 datasets, outperforming twelve state-of-the-art binary optimizers, including Binary Particle Swarm Optimization (bPSO), Binary Genetic Algorithm (bGA), and Binary Grey Wolf Optimizer (bGWO). The results demonstrate that DBSMOA delivers competitive accuracy, compact feature subsets, and strong computational efficiency, highlighting its robustness, scalability, and suitability for high-dimensional feature selection and other binary optimization tasks.