Improved Version of Kernelized Fuzzy C-Means using Credibility

Abstract

-Fuzzy c-means is a clustering algorithm which performs well with noiseless data-sets. Various disadvantages of FCM are its sensitivity towards noise points and able to detect only spherical clusters due to euclidean distance metric and can work with only linear data. Kernel approaches can improve the performance of conventional clustering. It changes the behavior of algorithm from linear separability to non-linear separability. It can be achieved by using kernel function as a distance metric, which transforms the data to higher dimensional space and find the difference between points considering all the characteristics of data which are not accessible in two dimensional space. Kernel fuzzy C-means (KFCM) algorithm can efficiently work with non-linear data. But still it is sensitive to noisy points. This paper proposed kernel credibilistic fuzzy C-means (KCFCM) algorithm that uses credibility to reduce the sensitivity of noisy points. Several experimental results show that the proposed algorithm can outperform other algorithms for general data with additive noise.