Boosting chameleon swarm algorithm with consumption AEO operator for global optimization and feature selection

摘要

Feature selection (FS) plays a crucial role as a pre-processing tool in data mining, especially for real-world applications in medical fields; it has been utilized exponentially and becomes very interesting to detect beneficial medical decisions. FS is deemed to be an NP-hard optimization problem that seeks to reduce feature size while maximizing the generalization of machine learning (ML) models. Chameleon swarm algorithm (CSA) is a recent devised metaheuristic algorithm inspired by chameleons’ intelligent behavior in nature. Because of its simplicity and ease of implementation, it has piqued the interest of numerous researchers to apply it in different fields. However, it suffers from premature convergence, the inadequate balance between exploration and exploitation, and being trapped in local optima. Accordingly, this paper develops an improved version of the CSA for FS, referred to as modified chameleon swarm algorithm (mCSA). This improvement was achieved by introducing three modifications to the original CSA to improve how well it performed. Firstly, we propose a non-linear transfer operator to achieve a better balance between exploration and exploitation. Secondly, we introduce a randomization Lévy flight control parameter to avoid stagnation and early convergence. Thirdly, we boost the global search strategy of the original CSA by the consumption operator of the Artificial ecosystem-based optimization (AEO) algorithm. The performance of the proposed mCSA method is assessed by employing the CEC2020 test suite for numerical optimization and fourteen UCI datasets, including real-world application (breast cancer diagnosis) for feature selection, and compared with several well-known algorithms and sophisticated approaches. The experimental results indicate that the proposed mCSA performs significantly better and outperforms existing comparative methods.