A novel discretization method for microarray-based cancer classification

Abstract

In this paper, we propose a gene expression diversity-based method for gene expression discretization. By counting the numbers of samples of different classes in an open expression intervals, the method calculates class distribution diversity and then expression diversity for genes. Based on the gene expression diversity, three discretization criteria are established for discretizing gene expression levels. We evaluate the proposed method on the publicly available leukemia dataset and compare it with several previous methods. © 2012 Springer-Verlag.