k-centroids clustering for asymmetric dissimilarities

In this paper, an asymmetric version of the kcentroids clustering algorithm is proposed. The asymmetry arises from the use of the asymmetric dissimilarities in the k-centroids algorithm. Application of the asymmetric measures of dissimilarity is motivated by the basic nature of the k-centroids algorithm, which uses dissimilarities in the asymmetric manner. It finds the minimal dissimilarity between an object being currently allocated, and one of the clusters centroids. Clusters centroids are treated as the dominant points governing the asymmetric relationships in the entire cluster analysis. The results of the experimental study on real and simulated data have shown the superiority of the asymmetric dissimilarities employed for the k-centroids method over their symmetric counterparts.