Minimization of the Number of Iterations in K-Medoids Clustering with Purity Algorithm
Rozzi Kesuma Dinata, Sujacka Retno, Novia Hasdyna
Abstract
With k-medoids algorithm, it often takes many iterations to cluster a large dataset, that is, the k-medoids algorithm cannot achieve the optimal performance. Based on cluster validity, this paper tries to optimize the clustering performance of k-medoids algorithm, using the purity algorithm. Specifically, the medoids value was determined by the purity value, and cluster validity was measured with the Davies Bouldin Index (DBI) on the Iris Dataset and the Death/Birth Rate Dataset. The results show that the cluster validity of the proposed purity k-medoids algorithm was better than the conventional k-medoids algorithm. The conventional k-medoids converged in an average of 8.7 iterations on the Death Birth Rate Dataset and 13.2 on the Iris Dataset. By contrast, the purity k-medoids algorithm only needed 2 iterations on either dataset. Therefore, the purity k-medoids algorithm can effectively minimize the number of iterations in the clustering of large datasets.