Cooperative Coevolutionary Probability-Based Binary Particle Swarm Optimization for High-Dimensional Feature Selection

Feature Selection (FS) on high-dimensional data is challenging for evolutionary computation-based FS algorithms. To address the vast decision space of high-dimensional FS problems, this paper proposes a Cooperative Coevolutionary Probability-based Binary Particle Swarm Optimization (CCPBPSO-FS) algorithm. The proposed algorithm divides the feature space into subspaces and employs multiple swarms in these subspaces to cooperatively solve the FS problem. A multi-swarm initialization strategy, considering feature ranking information, is proposed to initialize sparse particles. Additionally, a new probability-based solution updating mechanism with a mutation operator to enhance search performance is proposed. Experimental results on eight high-dimensional datasets demonstrate that CCPBPSO-FS achieves competitive classification performance while significantly reducing the number of selected features compared with benchmark FS algorithms.