Impact of Solution Representation in Nature-Inspired Algorithms for Feature Selection

Abstract

This paper investigates how does the solution representation in nature-inspired algorithms impact the performance of feature selection in classification problems. Four most suitable nature-inspired algorithms for feature selection were considered in the analysis, namely the Differential Evolution, Artificial Bee Colony, Particle Swarm Optimization, and Genetic Algorithm. The binary-coded and real-coded variants of the mentioned algorithms were compared for filter-based and wrapper-based feature selection methodologies on datasets commonly used by the research community. Additionally, the algorithms' performance on reducing the feature subset size regarding different solution representations was compared. Statistical tests were performed for discovering any significant differences in the algorithms' performances.