Distance-Based K-Medoids Clustering for Gene Expression Data

This paper proposes K-medoids algorithm for clustering gene expression data. It improves the selection of initial and updated medoids. Results of K-medoids clustering are mainly based on initial centroids; the proposed method helps find the best initial centroids. In this approach, distance matrix is found once which helps select the first k smallest objects as initial medoids. The main features of the algorithm are: (a) Distance matrix is calculated only once and used in the entire process. It avoids scanning of large database every time; and (b) No initial parameter requirement of k (number of clusters). Both these features help reduce execution time and the need for memory. The proposed algorithm is applied on synthetic, genome expression and medical datasets. The outcomes are validated using various measures like Rand index and FM index. Experiments show that the proposed algorithm runs faster and finds better results than the existing algorithms