An Enhanced Hybrid Binary Grey Wolf and Harris Hawk Optimization Algorithm Based on Cumulative Binomial Probability for Feature Selection in Classification

Abstract

Feature selection is a widely used approach for reducing dimensionality in datasets by eliminating irrelevant and redundant features. It significantly enhances the accuracy and efficacy of classification models. Hybrid binary grey wolf with Harris hawk optimization (HBGWOHHO) is a metaheuristic algorithm that has been effectively employed for feature selection in classification. However, the HBGWOHHO algorithm has a limitation in unbalanced exploration and exploitation in achieving the sub-optimal solution. This limitation refers to the linearly declining value of a balancing parameter, which lacks regulation between the exploration and exploitation phases. This paper presents an enhanced HBGWOHHO that employs an adaptive technique based on cumulative binomial probability (CBP) called hybrid grey wolf Harris hawk optimization-CBP (HBGWHHO_CBP) to fine-tune the balancing parameter. This adaptive adjustment technique ensures a more effective trade-off between exploration and exploitation, thus improving the algorithm's search efficiency and solution quality. Dimension-wise diversity metric is used to quantitatively assess this balance during the optimization process. Eleven UCI benchmark datasets were utilized to assess the efficacy of the proposed HBGWHHO_CBP. The proposed algorithm demonstrated superior performance across the evaluated datasets, yielding an average accuracy of 0.94, a mean of 8.51 selected features, and a mean fitness value of 0.06, while requiring less computational time. The Wilcoxon signed-rank test results indicate that the proposed algorithm significantly outperforms the native HBGWOHHO and three other metaheuristic-based feature selection algorithms. The proposed metaheuristic can be applied for addressing the feature selection in classification.