Combined three feature selection mechanisms with LVQ neural network for colon cancer diagnosis

Abstract

In this paper, a novel two-stage classification method for colon cancer diagnosis based on gene expression data is introduced, which combine three feature selection mechanisms with Learning Vector Quantization Neural Network (LVQNN). The first stage is to select effective informative gene based on Bhattacharyya distance, pairwise redundancy analysis (PRA) and principal component analysis (PCA) for dimension reduction and feature extraction. In the second stage, LVQ Neural Network is employed to construct a cancer data classifier. To show the validity of the method presented, the gene expression profile data set of colon cancer was used for classifying. The experimental results show that the proposed method can effectively identify colon cancer. Compared with other three neural network methods, LVQ Neural Network has the best classification effect. © 2011 Springer-Verlag.