M-Estimator induced Fuzzy Clustering Algorithms
Roland E. Winkler, Frank Klawonn, Rudolf Kruse · 2011
M-estimators can be seen as a special case of robust clustering algorithms.In this paper, we present the reversed direction and show that clustering algorithms can be constructed by using M-estimators.A clever normalization is used to link the values of several M-estimator prototypes together in one clustering algorithm.A variety of M-estimators and several normalization strategies are used in 4 data sets to present their differences and properties.The results are evaluated using 5 different clustering validation indices.