A noise-resistant fuzzy clustering approach with probabilistic typicalities

Probabilistic typicalities are different concepts comparing with probabilistic memberships in fuzzy clustering. They give another description of relation between data points and clustering centers. In terms of function optimization techniques, this paper presents a noise-resistant fuzzy clustering approach with probabilistic typicalities. Moreover, the usage of exponential functions greatly enlarges the relativity of points to centers based on the Euclidean distance. At last, its variety is also presented and behaves better on the IRIS data than the existed methods in computation precision. Experimental comparisons on noisy data and IRIS data show that our approach is hardly affected by noise and more accurate in computing the cluster centers than existed methods.