In recent years, hybrid feature selection methods incorporating global–local frameworks have gained significant attention due to their advantages. Leveraging the capabilities of swarm and evolutionary algorithms, we propose a modified Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) algorithm within a bi-level global–local search framework to achieve improved exploration while maintaining comparable execution time. In this method, we introduce a modified GA to perform the global search step using a subspace-wise search strategy. The primary motivation behind this approach is to reduce the search space, enabling the method to consider all parts of the space and avoid getting trapped in local optima. To enhance diversity without increasing complexity, we introduce a novel encoding scheme where each chromosome represents a subspace of solutions instead of a single solution. During each iteration, the PSO algorithm is employed to perform local search and identify high-quality solutions within each subspace. This step can be applied to subspaces in parallel to make the method faster. Notably, the PSO algorithm is applied only to specific dimensions of the individuals, resulting in significantly faster computation. Consequently, the solution selected by the PSO within each subspace serves as the basis for evaluating the fitness of the corresponding chromosome. Experiments conducted on 30 well-known datasets demonstrate that our method exhibits significantly better performance compared to state-of-the-art methods, both in low and high-dimensional data settings. Furthermore, our method demonstrates stability, making it suitable for real-world applications.
