Authors: Heba F. Eid
Addresses: Faculty of Science, Al-Azhar University, Cairo, 11759, Egypt
Abstract: Feature selection process is considered as one of the most difficult challenges in machine learning and has attracted many researchers recently. The main disadvantages of the classical optimisation algorithms based feature selection are slow convergence speed and local optima stagnation. In this work, a novel binary version of the whale optimisation is proposed for selecting the optimal feature subset and increasing the classification accuracy. The performance of the proposed binary whale optimisation (BWO) is verified by comparisons with three well known optimisation based feature selection algorithms; genetic algorithm, ant colony optimisation and particle swarm optimisation; on nine benchmark datasets. The qualitative and quantitative results show the capability of the proposed BWO to search the feature space for optimal feature combinations. Moreover, results prove that the proposed BWO is able to outperform the current algorithms on the majority of datasets in terms of both average classification accuracy and convergence speed.
Keywords: heuristic algorithm; bio-inspired optimisation; whale optimisation; BWO; binary whale optimisation; feature selection; PSO; particle swarm optimisation; genetic algorithm; ACO; ant colony optimisation.
International Journal of Metaheuristics, 2018 Vol.7 No.1, pp.67 - 79
Received: 25 May 2017
Accepted: 16 Sep 2017
Published online: 20 May 2018 *