A Framework of Gene Subset Selection Using Multiobjective Evolutionary Algorithm

Microarray gene expression technique can provide snap shots of gene expression levels of samples. This technique is promising to be used in clinical diagnosis and genomic pathology. However, the curse of dimensionality and other problems have been challenging researchers for a decade. Selecting a few discriminative genes is an important choice. But gene subset selection is a NP hard problem. This paper proposes an effective gene selection framework. This framework integrates gene filtering, sample selection, and multiobjective evolutionary algorithm (MOEA). We use MOEA to optimize four objective functions taking into account of class relevance, feature redundancy, classification performance, and the number of selected genes. Experimental comparison shows that the proposed approach is better than a well-known recursive feature elimination method in terms of classification performance and time complexity.