A combined MI-AVR approach for informative gene selection

Accurate classification of diseases from microarray gene expression profile is a challenging task because of its high dimensional low sample data. Most of the gene selection methods employ the criterion function on the entire microarray samples only once which cannot exactly represent the relevance among genes. This paper proposes a hybrid gene selection algorithm that selects genes in two stages, initially with all samples using Mutual Information followed by only unclassified samples using Augmented Variance Ratio. Feed Forward Neural Network trained by Back Propagation algorithm is used to classify the samples. The performance of the proposed approach is tested using six gene expression datasets. Simulation results show that the proposed method selects the genes which are highly informative and produces good classification accuracy than other methods reported in the literature.