Improved Binary Gray Wolf Optimizer Based on Adaptive β-Hill Climbing for Feature Selection

According to the literature reviews, the Gray Wolf Optimization (GWO) algorithm has been applied to various optimization problems, including feature selection. It is important to consider two opposing ideas while using the metaheuristic technique, exploring the search field, and exploiting the best possible solutions. Despite the increased performance of the GWO, stagnation in local optima areas could still be a concern. This paper proposes a hybridized version of Binary GWO (BGWO) and another recent metaheuristic algorithm, namely adaptive beta-hill climbing (A beta CH), to enhance the performance of a wrapper-based feature selection approach. The sigmoid transfer function is used to transfer the continuous search space into a binary version to meet the feature selection nature requirement. The K-Nearest Neighbor (KNN) classifier is used to evaluate the goodness of the selected features. To validate the performance of the proposed hybrid approach, 18 standard feature selection UCI benchmark datasets were used. The performance of the proposed hybrid approach was also compared with the Binary hybrid Gray Wolf Optimization Particle Swarm Optimization (BGWOPSO), BGWO (bGWO1,bGWO2), Binary Particle Swarm Optimization (BPSO), Binary Genetic Algorithm (BGA), Whale Optimization Algorithm with Simulated Annealing (WOASAT-2), A beta HC with Binary Sailfish (A beta BSF), Binary beta-Hill Climbing (beta HC), Binary JAYA with Adaptive Mutation (BJAM), and Binary Horse herd Optimization Algorithm(BHOA). The findings revealed that the proposed hybrid algorithm was effective in improving the performance of the normal BGWO algorithm, also the proposed hybrid approach outperforms the two approaches of the BGWO algorithm in terms of accuracy and selected feature size. Similarly, compared with BGWOPSO, BPSO, BGA, WOASAT-2, A beta BSF, beta HC, BJAM, and BHOA feature selection approaches, the proposed approach surpassed them and yielded better accuracy and smaller size of feature selection.