Multi-strategy fusion binary SHO guided by Pearson correlation coefficient for feature selection with cancer gene expression data

Abstract

Cancer gene expression data is extensively utilized to address the challenges of cancer subtype diagnosis. However, this data is often characterized by high-dimensional, multi-text and multi-classification, which requires an effective feature selection (FS) method. A multi-strategy fusion binary sea-horse optimizer guided by Pearson correlation coefficient was proposed for FS with cancer gene expression data. For the multi-strategy fusion, the rest strategy is introduced in the sea-horse motor behavior stage. Subsequently, a search strategy based on symbiotic organisms of sea horses is designed for the predation stage. Finally, the elementary function dynamic weight strategy is proposed. Multi-strategy fusion enables the sea-horse optimizer (SHO) to perform dynamic exploitation and exploration in the early stage of iteration, expand the search scope initially, and narrow the search scope in the middle and later stages of the algorithm, so as to avoid the algorithm falling into the local optimal and increase the possibility of the algorithm jumping out of the local optimal, and avoid the blind search caused by elite influence. In the FS part, Pearson correlation coefficient guided strategy is proposed firstly to add or delete features. Then eight binary algorithms are derived from S-type and V-type transfer functions. The simulation experiment was divided into four parts. Firstly, the CEC-2022 test functions were used to test the performance of the multi-strategy fusion SHO, from which the best variant TanASSHO was selected, and then compared with other nine swarm intelligent optimization algorithms. Performance tests of various algorithm variants on 18 UCI datasets show that V1PTASSHO is the most effective binary version. Finally, V1PTASSHO was compared with other nine swarm intelligent optimization algorithms on 18 cancer gene expression datasets. The results demonstrate that V1PTASSHO effectively reduces feature subsets, improve classification accuracy and obtain lower fitness value. Friedman test and Wilcoxon rank sum test were used for statistical analysis to verify the effectiveness of the proposed algorithm.