A STUDY ON GENE SELECTION AND CLASSIFICATION ALGORITHMS FOR CLASSIFICATION OF MICROARRAY GENE EXPRESSION DATA

The development of microarray technology allows researchers to monitor the expression of genes on a genomic scale. One of the main applications of microarray technology is the classification of tissue samples into tumor or normal tissue. Gene selection plays an important role prior to tissue classification. In this paper, a study on numerous combinations of gene selection techniques and classification algorithms for classification of microarray gene expression data is presented. The gene selection techniques include Fisher Criterion, Golub Signal-to-Noise, traditional t-test and Mann-Whitney rank sum statistic. The classification algorithms include support vector machines (SVMs) with several kernels and k-nearest neighbor (k-nn). The performance of the combined techniques is validated by using leave-one-out cross validation (LOOCV) technique and receiver operating characteristic (ROC) is used to analyze the results. The study demonstrated that selecting genes prior to tissue classification plays an important role for a better classification performance. The best combination is obtained by using Mann-Whitney Rank Sum Statistic and SVMs. The best ROC score achieved for this combination is at 0.91. This should be of significant value for diagnostic purposes as well as for guiding further exploration of the underlying biology.