A Comparative Study of Clustering Methods for Relevant Gene Selection in Microarray Data

Classification of microarray cancer data has drawn the attention of research community for better clinical diagnosis in last few years. Microarray datasets are characterized by high dimension and small sample size. Hence, the conventional wrapper methods for relevant gene selection cannot be applied directly on such datasets due to large computation time. In this paper, a two stage approach is proposed to determine a subset containing relevant and non redundant genes for better classification of microarray data. In first stage, genes were partitioned into distinct clusters to identify redundant genes. To determine the better choice of clustering algorithm to group redundant genes, four different clustering methods were investigated. Experiments on four well known cancer microarray datasets depicted that hierarchical agglomerative with complete link approach performed the best in terms of average classification accuracy for three datasets. Comparison with other state-of-art methods have shown that the proposed approach which involves gene clustering is effective in reducing redundancy among selected genes to provide better classification.