Fast and Efficient Feature Selection in AI Application Based on Enhanced Binary Secretary Bird Optimization Algorithm

Abstract

Metaheuristic algorithms, which draw inspiration from natural phenomena, have emerged as robust tools within computational intelligence and are widely applied across various fields. The effective use of artificial intelligence requires extracting pertinent information from extensive datasets. Working with big data presents several obstacles, including high dimensionality, duplicate data, and extraneous information. Feature selection techniques aim to reduce complexity by identifying and removing unnecessary attributes, which helps optimize computational resources in terms of both processing time and storage requirements. This paper introduces an enhanced binary variant of the Secretary Bird Optimization Algorithm (SBOA) designed to address feature selection challenges. The SBOA is a recent metaheuristic approach that replicates the survival tactics of secretary birds, specifically their hunting and predator avoidance behaviors. As computational methods, metaheuristic algorithms help solve complex optimization tasks. The proposed EB-SBOA incorporates two key improvements to the original SBOA: a refracted opposition-based learning method during initialization to expand population diversity, and a random replacement mechanism to improve convergence precision. The algorithm's effectiveness was tested using 25 benchmark datasets and compared against six contemporary wrapper-based feature selection techniques. Results demonstrate that EB-SBOA achieves superior performance in three key metrics: classification accuracy, average fitness value, and feature reduction. The findings' statistical validity was confirmed through Wilcoxon rank-sum testing.