Performance is more dependent on the number of clusters used than on the quality of the classification