Whale optimization approaches for wrapper feature selection

• Published in: Applied soft computing Vol. 62; pp. 441 - 453
• Main Authors: Mafarja, Majdi, Mirjalili, Seyedali
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.01.2018
• Subjects: Classification; Crossover; Evolutionary operators; Feature selection; Mutation; Optimization; Selection; Whale optimization algorithm; WOA; Feature selection; Selection; Classification; WOA; Evolutionary operators; Whale optimization algorithm; Mutation; Crossover; Optimization
• ISSN: 1568-4946; 1872-9681
• DOI: 10.1016/j.asoc.2017.11.006

[Display omitted] •Novel Whale Optimization approaches are proposed for feature Selection.•Crossover and mutation are used to enhance the exploitation property in WOA algorithm.•Tournament selection is used to enhance the exploration in WOA algorithm.•A superior performance of the proposed approaches is proved in the experiments. Classification accuracy highly dependents on the nature of the features in a dataset which may contain irrelevant or redundant data. The main aim of feature selection is to eliminate these types of features to enhance the classification accuracy. The wrapper feature selection model works on the feature set to reduce the number of features and improve the classification accuracy simultaneously. In this work, a new wrapper feature selection approach is proposed based on Whale Optimization Algorithm (WOA). WOA is a newly proposed algorithm that has not been systematically applied to feature selection problems yet. Two binary variants of the WOA algorithm are proposed to search the optimal feature subsets for classification purposes. In the first one, we aim to study the influence of using the Tournament and Roulette Wheel selection mechanisms instead of using a random operator in the searching process. In the second approach, crossover and mutation operators are used to enhance the exploitation of the WOA algorithm. The proposed methods are tested on standard benchmark datasets and then compared to three algorithms such as Particle Swarm Optimization (PSO), Genetic Algorithm (GA), the Ant Lion Optimizer (ALO), and five standard filter feature selection methods. The paper also considers an extensive study of the parameter setting for the proposed technique. The results show the efficiency of the proposed approaches in searching for the optimal feature subsets.