A Novel Rough Hypercuboid Method for Classifying Cancers Based on Gene Expression Profiles

Microarray data analysis based on gene expression profiles is attracting more and more attention from researchers for finding functional genes and for classifying diseases. Various available approaches for selecting features and for classification can be exploited to manipulate such data. However, fewer methods can be elegantly adapted to accomplish this purpose. The main challenge is that such microarray data always involve much more genes than samples, and the expression values of genes always vary in different experimental conditions. This hampers the utilization of conventional statistical methods. In this paper, we propose a novel rough hypercuboid approach for classifying cancers based on the rough set theory. The approach dynamically constructs implicit hypercuboids that involve minimum amounts of misclassified samples and consequently induces classifiers. Experimental results on some cancer gene expression data sets and the comparisons with some other methods show that the proposed method is a feasible way of classifying cancer tissues in applications.