Binary grey wolf optimizer with a novel population adaptation strategy for feature selection

Abstract

Feature selection is a fundamental pre-processing step in machine learning that aims to reduce the dimensionality of a dataset by selecting the most effective features from the original features. This process is regarded as a combinatorial optimization problem, and the grey wolf optimizer (GWO), a novel meta-heuristic algorithm, has gained popularity in feature selection due to its fast convergence speed and easy implementation. In this paper, an improved binary GWO algorithm incorporating a novel Population Adaptation strategy called PA-BGWO is proposed. The PA-BGWO takes into account the characteristics of the feature selection problem and designs three strategies. The proposed strategy includes an adaptive individual update procedure to enhance the exploitation ability and accelerate convergence speed, a head wolf fine-tuned mechanism to exert the impact on each independent feature of the objective function, and a filter-based method ReliefF for calculating feature weights with dynamically adjusted mutation probabilities based on the ranking features to effectively escape from local optima. Experimental comparisons with several state-of-the-art feature selection methods on 15 classification problems demonstrate that the proposed approach can select a small feature subset with higher classification accuracy in most cases.