Research on Selection Method of the Optimal Weighting Exponent and Clustering Number in Fuzzy C-means Algorithm
Jian Cui, Qiang Li, Jun Wang, Zong Da-wei · 2010
The Fuzzy C-Means (FCM) algorithm is commonly used for clustering. The weighting exponent tn and the clustering number C are the important parameters in FCM algorithm. Conventional fuzzy clustering method must use both of the two prespecified parameters, so in this paper we analyses the original algorithm and studies on the optimal selection methods of the m and c by introducing the fuzzy decision theory and the validity index Vkwonbasing on the geometric structure of the dataset. Experiment results with the IRIS dataset show that this algorithm can obtain the optimal weighting exponent m* and the optimal clustering number C*. Moreover, the fact that the best value scope of m achieved in practical applications indicates that the method is effective.