*Result*: A two-stage gene selection method for biomarker discovery from microarray data for cancer classification.

*Abstract The microarrays permit experts to monitor the gene profiling for thousands of genes across an array of cellular responses, phenotype, and circumstances. Selecting a tiny subset of discriminate genes (biomarkers) from high dimensional data is one of the most significant tasks in bioinformatics. In this article, we develop a new hybrid framework by combining CMIM and AGA called CMIMAGA that can help to determine the significant biomarkers from the gene expression data. In the proposed approach, CMIM applied as a filter which is easy to understand and filter out most of the meaningless genes, and wrapper method as adaptive genetic algorithm (AGA) is employed to select the highly discriminating genes for distinguishing of instances from the reduced datasets. The AGA method uses the classifiers as a fitness function to select the relevant genes and to classify the tumour and cancer correctly. The performance of proposed approach is evaluated over six widely used microarray datasets using three classifiers, namely Extreme Learning Machine (ELM), Support Vector Machine (SVM), and k-nearest neighbor (k-NN). The experimental results reveal that our approach with ELM achieves the goal of better classification accuracy with a minimum number of genes and outperform to other filter and wrapper approaches. Highlights • In this study, we estimated numerous gene selection techniques to determine the useful biomarkers, including conventional filter strategies and recently developed strategies intended to acquire highly discriminative genes. • Proposed a new hybrid strategy which combines filtration system and wrapper strategy. • Experiments performed on six sets of biological data for cancer classification. • The hybrid strategy can decrease the consequence of the over-fitting and achieved the better quality of solutions. [ABSTRACT FROM AUTHOR]*