An efficient approach for microarray data classification using filter wrapper hybrid approach

Abstract

Microarrays store gene expression data of each cell; thereby microarray contains thousands of features. Each row represents gene samples and each column represents conditions. In any classification task, selection of irrelevant or redundant features greatly reduces the performance of classifier. Therefore, selection of optimal number of significant features is a major challenge for any classification problem. Filter and wrapper approaches are mostly used for feature subset selection. Filters are computationally faster but wrapper approach is more efficient in terms of classification accuracy. This paper proposes a hybrid approach combining the filters and wrappers is proposed which takes the features from both the filters and the wrappers. In the initial step, a feature subset is selected using filters and the feature subset size is further reduced using the wrapper approach. The proposed method is tested on Colon Tumor, B-Cell Lymphoma (DLBCL) and Leukemia datasets. Simulation results show that the hybrid method achieved higher accuracy with a smaller feature subset when compared to the existing methods.