Dependence of Two Different Fuzzy Clustering Techniques on Random Initialization and a Comparison

In the recent past Kernelized Fuzzy C-Means clustering technique has earned popularity especially in the machine learning community. This technique has been derived from the conventional Fuzzy C-Means clustering technique of Bezdek by defining the vector norm with the Gaussian Radial Basic Function instead of a Euclidean distance. Subsequently the fuzzy cluster centroids and the partition matrix are defined using this new vector norm. In our present work we have tried to show the effect of random initialization of the membership values on the performances of both these techniques. In addition to this we have tried to show the variation of the performance of Kernelized Fuzzy C-Means clustering technique with different values of the adjustable parameter of its vector norm. Using Partition Coefficient and Clustering Entropy as validity indices we have tried to make a comparison of the performances of these two clustering techniques. Keywords— Kernelized Fuzzy C-Means Clustering Technique, Fuzzy C-Means Clustering Technique, Gaussian Radial Basic Function, Euclidean Distance, Partition Coefficient, Clustering Entropy.