A Hybrid Method of Feature Extraction for Tumor Classification Using Microarray Gene Expression Data

Abstract

Classification of disease phenotypes using microarray gene expression data faces a critical challenge due to its high dimensionality and small sample size nature. Hence there is a need to develop efficient dimension reduction techniques to improve the class prediction performance. In this paper we present a hybrid feature extraction method to combat the dimensionality problem by combining F-score statistics with autoregressive (AR) model. The F-score statistics preselect the discriminant genes from the raw microarray data and then this reduced set is modeled by the AR method to extract the relevant information. A low complexity radial basis function neural network (RBFNN) is also introduced to efficiently classify the microarray data. Exhaustive simulation study on six standard datasets shows the potentiality of the proposed method with the advantage of reduced computational complexity.