A study on regularization effects of fuzzified memberships in FCM clustering
Katsuhiro Honda, Yui Matsumoto, Akira Notsu, Hidetomo Ichihashi · 2011
FCM clustering is a fundamental technique for capturing intrinsic cluster structures of multivariate data sets. This paper presents a comparative study on the regularization effects of Fuzzy c-Means memberships estimated by two different fuzzification approaches: standard approach and entropy regularization approach. In this paper, the characteristics of the two fuzzification approaches are also discussed in noise fuzzy clustering (NFC) and it is revealed that the noise rejection mechanism of NFC can contribute to weakening the influence of initialization problems in entropy regularization approach although the approach is generally more sensitive to initial partition than the standard approach.