Gene selection based on recursive spider wasp optimizer guided by marine predators algorithm

Abstract Detecting tumors using gene analysis in microarray data is a critical area of research in artificial intelligence and bioinformatics. However, due to the large number of genes compared to observations, feature selection is a central process in microarray analysis. While various gene selection methods have been developed to select the most relevant genes, these methods’ efficiency and reliability can be improved. This paper proposes a new two-phase gene selection method that combines the ReliefF filter method with a novel version of the spider wasp optimizer (SWO) called RSWO-MPA. In the first phase, the ReliefF filter method is utilized to reduce the number of genes to a reasonable number. In the second phase, RSWO-MPA applies a recursive spider wasp optimizer guided by the marine predators algorithm (MPA) to select the most informative genes from the previously selected ones. The MPA is used in the initialization step of recursive SWO to narrow down the search space to the most relevant and accurate genes. The proposed RSWO-MPA has been implemented and validated through extensive experimentation using eight microarray gene expression datasets. The enhanced RSWO-MPA is compared with seven widely used and recently developed meta-heuristic algorithms, including Kepler optimization algorithm (KOA), marine predators algorithm (MPA), social ski-driver optimization (SSD), whale optimization algorithm (WOA), Harris hawks optimization (HHO), artificial bee colony (ABC) algorithm, and original SWO. The experimental results demonstrate that the developed method yields the highest accuracy, selects fewer features, and exhibits more stability than other compared algorithms and cutting-edge methods for all the datasets used. Specifically, it achieved an accuracy of 100.00%, 94.51%, 98.13%, 95.63%, 100.00%, 100.00%, 92.97%, and 100.00% for Yeoh, West, Chiaretti, Burcyznski, leukemia, ovarian cancer, central nervous system, and SRBCT datasets, respectively.