Dynamic Swordfish Movement Optimization Algorithm for Feature Selection

Feature selection represents a crucial preprocessing step in machine learning pipelines, particularly when dealing with high-dimensional datasets that often contain redundant or irrelevant information. To address the challenge of efficiently selecting informative features while optimizing classification accuracy, this paper introduces the Dynamic Binary Swordfish Movement Optimization Algorithm (DBSMOA), a novel binary metaheuristic inspired by the foraging 1 behavior of swordfish. DBSMOA enhances the original Swordfish Movement Optimization Algorithm (SMOA) by incorporating dynamic behavioral mechanisms that adaptively balance exploration and exploitation throughout the search process. The proposed algorithm employs a binary encoding strategy based on sigmoid probabilistic mapping and dynamically alternates agent roles using real-time performance metrics and elitist selection strategies. To assess its efficacy, DBSMOA is extensively evaluated on diverse benchmark datasets and compared against 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 consistently achieves superior performance in classification accuracy, feature reduction, and computational efficiency. These findings highlight the robustness and adaptability of DBSMOA for binary optimization problems, as well as its practical potential for high-dimensional data analytics.