Affinity Propagation Based Algorithm for Optimal K-Means Clustering

K-means clustering is widely used due to its fast convergence, but it is sensitive to the initial condition. The limitation of k-means algorithm is that the user has to specify the number of clusters (K). There are some methods to initialize the number of clusters. But those methods perform worse in some cases. So we are proposing a method called affinity propagation in this paper which resolves those problems. By making use of the convergence property of K-means and the good performance of affinity propagation, we presented a new clustering strategy which can produce much lower squared error than AP and standard k-means. The efficiency and effectiveness of our method is demonstrated through extensive comparisons with other methods using UCI datasets of high dimensionality.