Multiswarm heterogeneous binary PSO using win-win approach for improved feature selection in liver and kidney disease diagnosis

Feature selection is a significant preprocessing method in the classification part of an expert system. We propose a new Multiswarm Heterogeneous Binary Particle Swarm Optimization algorithm (MHBPSO) using a Win-Win approach to improve the performance of Binary Particle Swarm Optimization algorithm (BPSO) for feature selection. MHBPSO is a cooperation algorithm, which includes BPSO and its three variants such as Boolean PSO (BoPSO), Self Adjusted Hierarchical Boolean PSO (SAHBoPSO), and Catfish Self Adjusted Hierarchical Boolean PSO (CSAHBoPSO). It performs heterogeneous search on the entire solution space using four different algorithms. Each algorithm shares global best information to another algorithm to select the preeminent global best position. A variant of BoPSO, SAHBoPSO is proposed, in which leaders are identified based on the fitness values for guiding remaining particles and thus forms the hierarchical structure of leaders and its followers. Meanwhile leaders and followers are changed dynamically and consequently changed the hierarchical structure. CSAHBoPSO, which is the version of SAHBoPSO, is also proposed to avoid stagnation in the subsequent iterations. To assess the performance of the proposed algorithms CEC 2013 benchmark functions are employed. In order to validate the proposed algorithms, comparative study with BPSO, BoPSO, VPSO (Mirjalili and Lewis, 2013), HBPSOGA (Wang et al., 2018) and CatfishBPSO (Chuang et al., 2011a) is provided. Experimental results show that SAHBoPSO and CSAHBoPSO algorithm based on BoPSO are promising and significantly better than BPSO, BoPSO, and VPSO. MHBPSO shows the superior improvement in the search ability. In addition, proposed algorithms are tested in the feature selection phase of intelligent liver and kidney cancer diagnostic systems to select elite features from the liver and kidney cancer data. Findings show that the proposed system is proficient in selecting the elite features to classify the tumor as benign or malignant with minimum error rates. (C) 2018 Elsevier Ltd. All rights reserved.