Clustering-based hybrid feature selection approach for high dimensional microarray data

Abstract

The DNA microarrays are used to monitor the expression levels of significant genes. Most of the microarray data are assumed to be high dimensional, redundant, and noisy. This paper proposed a clustering-based hybrid gene selection approach to reduce the high dimensionality and increase the classification accuracy of cancer microarray data. The proposed approach uses the combined method of k-means clustering algorithm and signal-to-noise-ratio ranking method as a primary filtering method to reduce the high dimensionality of the microarray dataset. A cellular learning automaton combined with ant colony optimization is then applied on the reduced dataset as a wrapper method to get the optimized gene subset. The classifiers adopted to evaluate the proposed method are support vector machine, K-nearest neighbor, and Naive Bayes. The experiments showed promising results in gene subset selection and classification.