Research On Feature Gene Selection Method Based On Genetic Algorithm

With the completion of human genome project, life sciences have en tered a newera. Using microarray technologies，biologists can high through analysis thousandsof gene expression values．Feature gene selection aims to find a compact featuresubset used to construct a pattern classifier with reduced complexity, in order toimprove the classification performance. It is not only for us to find disease-relatedgenes and improve classification of tumors, but also reduces the cost of the clinicaldiagnosis of tumor type. Consequently，how to analysis and process this kind of dataand find meaningful gene subset has became a key factor for disease research andtreatment.An effective feature gene selection method should not only be able to produce asolution with better classification performance, but also it should have goodrobustness. Gene expression microarray data with characteristics of significantlyless sample and high dimension, some related studies confirm this kind of datasetmore easily lead to poor robustness of feature selection methods. However, theexisting feature selection methods are mostly concerned about the classificationperformance of the algorithm, easy to overlook the robustness of the algorithm.To compromise the robustness and prediction of gene selection, we propose amethod based on genetic algorithm. In this method, we firstly use cumulativedeviation to evaluate abnormal genes to avoid reducing performance of downstreamstudy. Then, to elevate robustness, we design a weight score method to fuse multiplecriteria. It not only considers complement between criteria, but also relativeimportance. More importantly, we only use multiple criteria to filter genes, but notrank them, to avoid possible wrong rank. Finally, we use genetic algorithm to searchoptimum gene subset in feature combinatorial. The experimental results show that itget better prediction accuracy and robustness.