Binary Owl Search Algorithm for Feature Subset Selection

Abstract

Feature subset selection is one of the essential preprocessing tasks for numerous classification problems. This is because good feature subset can reduce overfitting of data, enhance the accuracy, and lessen the training time of a classifier model. However, finding the optimum feature subset is computationally expensive when the number of features is relatively high. Therefore, stochastic approaches are often used in attaining good feature subset within a feasible time frame. In this paper, we propose a binary variant of recent stochastic optimization algorithm Owl Search Optimization (OSA) for optimum feature subset selection. In this approach, six different transfer functions of S-shaped and V-shaped families were employed for generating six different binary Owl Search Optimization (BOSA) models. The proposed mechanism then employed on eleven publicly available datasets and performances were compared with popular approaches including, Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Harmony Search (HC). Results reveal that, for most of the datasets, BOSA-based approaches can produce optimal feature subset with reduced number of features and improved classification accuracy compared to other approaches.