Binary Puma Optimizer: A Metaheuristic Approach for Gene Selection in Bioinformatics with Machine Learning Classification

Abstract Gene expression data is a matrix derived from DNA microarray analysis; a single chip can contain a thousand genetic instructions via DNA microarray technology. Each microarray experiment can simultaneously evaluate thousands of genes. Algorithms for pattern recognition are commonly employed to differentiate between samples of healthy and malignant patients using gene expression data. Gene expression data is categorized as high-dimensional data and typically includes irrelevant, noisy, and redundant genes. Such datasets confront the challenge of the application of machine learning algorithms. This is due to their requirement for extensive computational resources, which negatively impacts classification performance and obstructs the training and testing processes. Gene selection is necessary to avoid these risks. This work proposes a unique binary puma optimizer (BPo) to choose the top-ranked genes. The modified Bpo uses a U-shaped transfer function to transform the algorithm from continuous to discrete. Nine widely utilized microarray datasets are employed to evaluate the performance of the proposed technique. The Support Vector Machine (SVM) is employed to evaluate the efficacy of the selected gene subset for the classification task and as a fitness function. The effectiveness of BPo is evaluated by achieving the least number of genes and the least time. Furthermore, utilizing the identical datasets, the results of BPo are compared with ten state-of-the-art gene selection methods based on classification accuracy and the number of genes selected. The results indicate that the proposed method provides the optimal results for the nine datasets. The experiment's results demonstrated the efficacy of the proposed method in exploring the gene search space while determining the optimal gene pairs. To statistically support the data, we performed a statistical analysis employing the Friedman and Wilcoxon tests, highlighting the significance of the proposed technique in comparison to existing algorithms based on accuracy. This analysis indicates that the proposed BPo substantially improves gene selection performance.