A novel chaotic selfish herd optimizer for global optimization and feature selection

Abstract

Selfish Herd Optimizer (SHO) is a recently proposed population-based metaheuristic inspired by the predatory interactions of herd and predators. It has been proved that SHO can provide competitive results in comparison to other well-known metaheuristics on various optimization problems. Like other metaheuristic algorithms, the main problem faced by the SHO is that it may easily get trapped into local optimal solutions, creating low precision and slow convergence speeds. Therefore, in order to enhance the global convergence speeds, and to obtain better performance, chaotic search have been augmented to searching process of SHO. Various chaotic maps were considered in the proposed Chaotic Selfish Herd Optimizer (CSHO) algorithm in order to replace the value of survival parameter of each searching agent which assists in controlling both exploration and exploitation. The performance of the proposed CSHO is compared with recent high performing meta-heuristics on 13 benchmark functions having unimodal and multimodal properties. Additionally the performance of CSHO as a feature selection approach is compared with various state-of-the-art feature selection approaches. The simulation results demonstrated that the chaotic maps (especially tent map) are able to significantly boost the performance of SHO. Moreover, the results clearly indicated the competency of CSHO in achieving the optimal feature subset by accomplishing maximum accuracy and a minimum number of features.