Hybrid binary COOT algorithm with simulated annealing for feature selection in high-dimensional microarray data

Microarray analysis of gene expression can help with disease and cancer diagnosis and prognosis. Identification of gene biomarkers is one of the most difficult issues in microarray cancer classification due to the diverse complexity of different cancers and the high dimensionality of data. In this paper, a new gene selection strategy based on the binary COOT (BCOOT) optimization algorithm is proposed. The COOT algorithm is a newly proposed optimizer whose ability to solve gene selection problems has yet to be explored. Three binary variants of the COOT algorithm are suggested to search for the targeting genes to classify cancer and diseases. The proposed algorithms are BCOOT, BCOOT-C, and BCOOT-CSA. In the first method, a hyperbolic tangent transfer function is used to convert the continuous version of the COOT algorithm to binary. In the second approach, a crossover operator (C) is used to improve the global search of the BCOOT algorithm. In the third method, BCOOT-C is hybridized with simulated annealing (SA) to boost the algorithm's local exploitation capabilities in order to find robust and stable informative genes. Furthermore, minimum redundancy maximum relevance (mRMR) is used as a prefiltering technique to eliminate redundant genes. The proposed algorithms are tested on ten well-known microarray datasets and then compared to other powerful optimization algorithms, and recent state-of-the-art gene selection techniques. The experimental results demonstrate that the BCOOT-CSA approach surpasses BCOOT and BCOOT-C and outperforms other techniques in terms of prediction accuracy and the number of selected genes in most cases.