Two-stage clustering using one-pass K-medoids and medoid-based agglomerative hierarchical algorithms

Clustering on weighted networks needs consideration of particular techniques. Medoid-based clustering seems promising in this sense. Although K-medoids have already been studied, agglomerative hierarchical clustering using medoids should also be studied. A drawback of agglomerative hierarchical clustering is that large number of objects cannot be handled, while it has the advantage of having much information on the process of cluster generation. We propose a two-stage method of clustering that has the advantage of agglomerative hierarchical clustering and handles a large number of objects. In the first stage a one-pass K-medoids++ is used to have a medium number of small clusters, and then medoid-based agglomerative hierarchical procedure is applied. Numerical examples show effectiveness of the proposed method.