Abstract:

In order to overcome the weakness of present ensemble methods for fuzzy clustering, a novel method of fuzzy clustering ensemble is proposed. In the framework of evidence ensemble, the recognition framework, basic probability assignment and synthesis rules are analyzed. Three different basic probability assignment (BPA) methods is proposed, which are approximation probability of clusters, normalized fuzzy Hamming distance, and evidence synthesis of the two. Because of complementary advantages of the two methods, the third method is able to get more useful BPA. At last, parameter settings, convergence and effectiveness of the method proposed are analyzed by experiment.