An optimized feature selection using triangle mutation rule and restart strategy in enhanced slime mould algorithm

Abstract

This paper proposes an improved feature selection method based on an improved Slime Mould Algorithm (SMA), called the Triangular Mutation Rule Restart Strategy Slime Mould Algorithm (TRSMA), to overcome some of the shortcomings of the SMA, including premature convergence, poor population diversity, and local optima entrapment. The TRSMA uses three main strategies: (1) Good Point Set (GPS) initialization for better initial population diversity, (2) Triangular Mutation Rule (TMR) for better global exploration and finding higher-quality areas in the solution space, and (3) a Restart Strategy (RS) to reinitialize weak individuals to escape from local optimum. Then we combine the TRSMA with Support Vector Machines (SVM) and propose the TRSMA-SVM model to select the joint feature and classifier parameters. Experimental results on nine University of California, Irvine (UCI) datasets and a real-world malaria dataset show that TRSMA-SVM consistently outperforms recent state-of-the-art methods regarding classification accuracy with fewer selected features. Comparison with benchmark testing on CEC2017 functions confirms TRSMA’s ability to perform global optimization. Statistical tests using the Wilcoxon rank-sum and Friedman tests also verify these performance gains. The results illustrate that TRSMA is powerful and can handle complex high-dimensional optimization problems.