Feature Selection in Microarray using Proposed Hybrid Minimum Redundancy-Maximum Relevance (MRMR) and Modified Genetic Algorithm (MGA)

Abstract

Gene expression microarray data commonly have an enormous number of genes with a smaller number of samples. In these genes, many are irrelevant, insignificant or redundant for the classification analysis. Therefore, the identification of informative genes, which have the greatest role in classification and diagnosis, is of essential and practical importance to the classification problems, such as cancer versus non-cancer classification and classification of different tumor types. This paper aims to present a novel idea for implementing MRMR, the hybrid Minimum Redundancy-Maximum Relevance method combined with a Modified Genetic Algorithm, to minimize the selection of microarray data feature sets. This paper proposes a two-step feature selection algorithm by integrating Minimum Redundancy Maximum Relevance (MRMR) and Modified Genetic Algorithm (MGA). In the first step, MRMR is used to filter redundant genes in high-dimensional microarray data. The second step is used to eliminate irrelevant genes. The proposed MRMR-MGA algorithm is compared with traditional MRMR with the GA algorithm. The implementation results show that the proposed method has good selection and classification performances.