A Comparative Study on Regularization Effects of FCM-type Fuzzy Clustering

Abstract

The FCM-type clustering is the fuzzy extension of k-Means, in which membership fuzzification is achieved by such approaches as the standard model with membership exponents or the regularization model with penalty terms (entropy terms). This research compares the fuzzification approaches from the view point of regularization effects of objective functions. In addition, considering the difference of fuzzy classification functions, the initialization effects are also discussed in noise clustering extensions.