An ensemble soft weighted gene selection-based approach and cancer classification using modified metaheuristic learning

Hybrid algorithms are effective methods for solving optimization problems that rarely have been used in the gene selection procedure. This paper introduces a novel modified model for microarray data classification using an optimized gene subset selection method. The proposed approach consists of ensemble feature selection based on wrapper methods using five criteria, which reduces the data dimensions and time complexity. Five feature ranking procedures, including receiver operating characteristic curve, two-sample T-test, Wilcoxon, Bhattacharyya distance, and entropy, are used in the soft weighting method. Besides, we proposed a classification method that used the support vector machine (SVM) and metaheuristic algorithm. The optimization of the SVM hyper-parameters for the radial basis function (RBF) kernel function is performed using a modified Water Cycle Algorithm (mWCA). The results indicate that the ensemble performance of genes-mWCA SVM (EGmWS) is considered an efficient method compared to similar approaches in terms of accuracy and solving the uncertainty problem. Five benchmark microarray datasets, including leukemia, MicroRNA-Breast, diffuse large B-cell lymphoma, prostate, and colon, are employed for experiments. The highest and lowest numbers of genes are related to prostate with 12 533 genes and MicroRNA-Breast with 1926 genes, respectively. Besides, the highest and lowest numbers of samples are MicroRNA-Breast with 132 samples and colon with 62 samples, respectively. The results of classifying all data by applying effective genes of the EF-WS yielded high accuracies in microarray data classification. In addition to the robustness and simplicity of the proposed method, the model's generalizability is another crucial aspect of the method that can be further developed to increase the accuracy while reducing classification error.

Automated summary

Hybrid algorithms, feature selection methods, and classification algorithms are combined to improve the efficiency and accuracy of microarray data classification, demonstrating promising results on five benchmark datasets.