Authorization

 

Journal

Volume 10, №1, Special Issue on Fuzzy Set Theory and Applications

Fuzzy clustering with polynomial fuzzifier function in connection with M-estimators (pages 146-163)
Roland Winkler, Frank Klawonn, Rudolf Kruse


Abstract. The invention of fuzzy set theory by Lotfi A. Zadeh had great impact on linguistic computing and machine controlling. Also data mining algorithms were inspired by this idea, routing back to fuzzy clustering. In this paper, we will explore the connections of fuzzy clustering approaches to hard clustering approaches as well as on statistically motivated algorithms. We will use Klawonn and Höppners idea of connecting hard clustering with fuzzy clustering and specialize their algorithms to define an M-estimator. Furthermore we investigate the transition from M-estimators to fuzzy clustering algorithms. To extend M-estimators to a multi-cluster problem, we use a clever normalization function of the robust weights, similar to fuzzy c-means. The resulting algorithm provides an update strategy that can be used with every robust loss function that can be used for M-estimators. However, the success of the clustering algorithm depends on the properties of the loss function.


Keywords:
Fuzzy c-means, M-estimators, Polynomial Fuzzifier Function, Robust Statistics, Noise Clustering, Multiple Prototypes, Robust Statistics.


AMS Subject Classification:
62H30.



COIA 2027

7-9 July 2027
Baku, Azerbaijan

Search