Binary atom search optimisation approaches for feature selection
Jingwei Too, Abdul Rahim Abdullah
Abstract
Atom Search Optimisation (ASO) is a recently proposed metaheuristic algorithm that has proved to work effectively on several benchmark tests. In this paper, we propose the binary variants of atom search optimisation (BASO) for wrapper feature selection. In the proposed scheme, eight transfer functions from S-shaped and V-shaped families are used to convert the continuous ASO into the binary version. The proposed BASO approaches are employed to select a subset of significant features for efficient classification. Twenty-two well-known benchmark datasets acquired from the UCI machine learning repository are used for performance validation. In the experiment, the BASO with an optimal transfer function that contributes to the best classification performance is presented. The particle swarm optimisation (PSO), binary differential evolution (BDE), binary bat algorithm (BBA), binary flower pollination algorithm (BFPA), and binary salp swarm algorithm (BSSA) are used to evaluate the efficacy and efficiency of proposed approaches in feature selection. Our experimental results reveal the superiority of proposed BASO not only in high prediction accuracy but also in the minimal number of selected features.