A Neighborhood Based Particle Swarm Optimization with Sine Co-sine Mutation Operator for Feature Selection

Abstract

Feature selection is a vital data pre-processing process in many practical applications. Feature selection aims to get rid of those unnecessary features and improve the performance of the classification model. In this paper, a neighborhood based particle swarm optimization with sine cosine mutation operator (NPSOSC) is proposed to select the most informative feature subset. The improvements are included to strengthen its search capacity and avoid local optima stagnation. A distance and fitness based neighborhood search strategy is developed to form stable neighborhood structures for the particles. Each particle adopts superior information from its neighborhoods and the entire swarm can search different regions of the entire search space. The second improvement incorporates a sine cosine mutation operator to enhance the exploration ability and add more randomness into the search process. The improvements will lead to an enhanced balance between exploration and exploitation. To demonstrate the performance of the proposed NPSOSC, seven well-known optimizers are compared with the NPSOSC on 16 well-regarded datasets with different difficulty levels. The experimental results and statistical tests demonstrate the excellent performance of the proposed NPSOSC in exploring the feature space and selecting the most informative features.