Performance enhancement of K-means clustering algorithm for gene expression data using entropy-based centroid selection

Micro array data play a vital role in simultaneously monitoring the expression profile of large number of genes that are specified with various experimental conditions. In bioinformatics research, the recognition of co-expressed and coherent patterns is a major objective in micro array data analysis. The K-means clustering algorithm is gaining popularity in the knowledge discovery domain for effectively analyzing this data by comparing gene expression profiles or sample expression profiles. However, the strategy employed in this algorithm is computationally expensive as regard to time complexity and selection of initial centroids. It selects initial centroids randomly that affect the quality of resulting clusters. In order to successfully deal this problem, we propose a variant method for finding initial centroids by using entropy-based farthest neighbor approach. Our experimental results illustrate the accuracy of gene clusters with less number of iterations in contrast to the traditional K-means clustering.