Kernel k'-means algorithm for clustering analysis

k'-means algorithm is a new improvement of k-means algorithm. It implements a rewarding and penalizing competitive learning mechanism into the k-means paradigm such that the number of clusters can be automatically determined for a given dataset. This paper further proposes the kernelized versions of k'-means algorithms with four different discrepancy metrics. It is demonstrated by the experiments on both synthetic and real-world datasets that these kernel k'-means algorithms can automatically detect the number of actual clusters in a dataset, with a classification accuracy rate being considerably better than those of the corresponding k'-means algorithms. ? 2013 Springer-Verlag.EI