IBBA: an improved binary bat algorithm for solving low and high-dimensional feature selection problems

Technological advancements have resulted in the accumulation of vast amounts of data across various industries, often containing redundant or irrelevant features. As a result, the development of efficient feature selection methods has become increasingly critical. This paper proposes an Improved Binary Bat Algorithm (IBBA) to overcome the limitations of the original Bat Algorithm (BA), particularly its weak exploration ability and tendency to become trapped in local optima. IBBA enhances both exploration and exploitation through a novel Fitness-based Exploitation Strategy (FES) and an improved Harris Hawks Optimization (HHO). Additionally, random perturbations are introduced during iterations to adjust positions that deviate from the search space, thus preventing ineffective searches. Since the original BA is primarily designed for continuous optimization problems, this study also investigates the effect of four V-shaped transfer functions on the algorithm's performance. Experimental results on 28 datasets with varying dimensionalities (ranging from nine to 12,600 features) demonstrate that IBBA outperforms 12 state-of-the-art metaheuristic algorithms in terms of fitness, accuracy, feature selection ratio, and runtime. Moreover, an analysis of exploration and exploitation shows that IBBA effectively balances these two processes, addressing BA's exploration shortcomings. The Wilcoxon signed-rank test, conducted at a significance level of 0.05, validates the algorithm's effectiveness, revealing that IBBA demonstrates significant advantages in 87.5% of the tests. Finally, comparisons with 14 recently proposed feature selection methods highlight IBBA's competitive classification accuracy. Therefore, this study is expected to make a valuable contribution to solving feature selection problems across datasets with diverse dimensionalities.