A comparative investigation of k-means and partition around Medoid methods of clustering : a case study with acute lymphoblastic leukemia data

Abstract EN

Clustering methods arc important tool in data mining.

The main challenge of clustering is to select the suitable method to be used for a given data set and the estimation of the number of clusters in the data set, especially in case of the unsuperviscd data.

In this paper, a comparison between two important partitioning clustering methods namely the K-means and the Partition Around Medoid (PAM) have been considered and a special index for each has been used to estimate number of clusters.

Also different indices of internal validation and stability measures have been used to compare these two methods to evaluate their performance by using these indices.

Internal validation and stability measures have been used to compare between AT-means and PAM for B-cclls and T-cells and it has been found that for B-cells the K-means performs better than PAM by Connectivity, Dunn, Silhouette, APN, ADM, FOM indexes and PAM perform better than K-means by AD index.

For T-cells, PAM performs better than K-means by Connectivity index and A^-means performs better than PAM by Dunn, Silhouette, APN, AD, ADM, FOM indices.

Keywords: B-cells, T-cells, K- means, PAM, Calinski, Silhouette, Connectivity, Dunn, APN, AD, ADM, FOM.