Hybrid metaheuristic algorithms for feature selection in classification: a systematic literature review

Abstract

Feature selection (FS) is a popular technique for improving machine learning (ML) model's effectiveness by eliminating irrelevant and redundant features. It is challenging because of the intricate relationship between features and large search space. Recent studies have focused on using hybrid metaheuristics to solve FS problem. This systematic literature review (SLR) is performed on three significant databases that explores recent studies from 2019 to 2024 that used hybrid metaheuristics for FS in classification. This paper aims to understand the existing hybrid algorithms, hybridization goal, hybridization type, and application domains. Moreover, crucial parameters, fitness and transfer functions, initial population method, traditional FS approach, classification algorithm, evaluation criteria, and statistical test are investigated in this paper. The qualitative findings derived from the systematic review encompassed 646 publications, systematically categorized based on predefined inclusion and exclusion criteria. Consequently, 35 papers were analyzed to develop new insights in the domain of FS in classification, focusing on single-objective metaheuristics. Hybrid metaheuristics surpass the efficacy of their individual components in enhancing algorithmic performance to attain optimal or near-optimal solutions. The limitations of hybrid metaheuristics and research gaps are identified for scholars interested in developing metaheuristic algorithms for FS.