The last step in training phase is refinement of the clusters found above. Although DynamicClustering counters all the basic k-means disadvantages, setting the intra-cluster similarity r may require experimentation. Also, a cluster may have a