Modified salp swarm algorithm based on competition mechanism and variable shifted windows for feature selection

Feature selection (FS) is used to reduce the dimensionality of datasets, which employs the most informative features to obtain the maximum classification accuracy. The swarm intelligent (SI) algorithm based FS method meets great challenges in initial population quality and search capability. On this account, this paper introduces a modified salp swarm algorithm based on competition mechanism and variable shifted windows called CVSSA. First of all, a competition mechanism is proposed to take full advantage of the characteristics of the random method and maximal information coefficient (MIC) based method to generate a high-quality initial population. Furthermore, the variable shifted windows strategy is introduced to control the numbers of features that enter into the position update to improve the exploitation capability of the algorithm. To comprehensively enhance the search performance of the algorithm, an improved movement mathematical model is designed. Last but not least, an adaptive generalized opposition-based learning (AGOBL) is introduced to further improve the exploitation capability and accelerate the convergence rate. A series of typical and state-of-the-art algorithms are used to make comparisons with the proposed CVSSA on some typical datasets. The experimental results reveal that the CVSSA provides better results than the other algorithms in key indexes. Meanwhile, Wilcoxon’s statistical test results establish that the advantage of the proposed algorithm is significant. It is established that the CVSSA is an efficient algorithm for the FS problems.