Computing and Informatics

With the advent of microarray technology, it is possible to measure gene expression levels of thousands of genes simultaneously. This helps us diagnose and classify some particular cancers directly using DNA microarray. High-dimensionality and small sample size of microarray datasets has made the task of classification difficult. These datasets contain a large number of redundant and irrelevant genes. For efficient classification of samples there is a need of selecting a smaller set of relevant and non-redundant genes. In this paper, we have proposed a two stage algorithm for finding a set of discriminatory genes responsible for classification of high dimensional microarray datasets. In the first stage redundancy is reduced by grouping correlated genes into clusters and selecting a representative gene from each cluster. Maximal information compression index is used to measure similarity between genes. In the second stage a wrapper based forward feature selection method is used to obtain a set of discriminatory genes for a given classifier. We have investigated three different techniques for clustering and four classifiers in our experiments. The proposed algorithm is tested on six well known publicly available datasets. Comparison with the other state-of-the-art methods show that our proposed algorithm is able to achieve better classification accuracy with less number of genes.