Horizontal crossover and co-operative hunting-based Whale Optimization Algorithm for feature selection

Abstract

The Whale Optimization Algorithm (WOA) is a widely-used approach for problem-solving, but it has some inherent limitations such as poor exploration capabilities, susceptibility to local optima, and reduced solution accuracy. To address these drawbacks, this study introduces a novel approach known as Horizontal Crossover and Co-operative-hunting-based WOA (HCCWOA). This enhanced algorithm incorporates a weight, co-operative learning techniques, and a horizontal crossover strategy into the WOA framework. The introduction of horizontal crossover bolsters the exploration capabilities of WOA, while the integration of co-operative learning techniques and an inertia weight enhances its exploitation abilities. Recognizing the significance of feature selection in this context, the proposed algorithm is applied in a wrapper mode with the K-Nearest Neighbor (KNN) classifier to select relevant features. The effectiveness of the HCCWOA is rigorously evaluated on twelve classical datasets sourced from the UCI repository. A comprehensive comparison is conducted with eight well-established metaheuristic algorithms and five recent variations of WOA. Performance metrics, including maximum accuracy, minimum fitness, and minimum feature count, are considered in this comparative analysis. The simulation results affirm that the HCCWOA outperforms other algorithms in at least six out of the twelve datasets. This enhanced performance is further substantiated through statistical analyses, including Friedman's rank test, paired-sample Wilcoxon signed-rank test, two-way ANOVA test, T-test, and boxplot analysis. The combination of empirical results and statistical validation supports the superior effectiveness of the proposed HCCWOA approach, highlighting its ability to effectively explore feature spaces and select the most relevant characteristics for classification tasks.