Clustering With Kernel K-means And Diffusion Distance

K-means is a traditional algorithm for clustering problem. Kernel K-means is analternative of K-means algorithm with replacing “Distance” by “Kernel”. This replace-ment will alter the dimension structure of data,therefore cut data nonlinearly. KernelK-means behaves better thani K-meanswhen applied on such nonlinear data.Diffusion maps would natrally lead the definition of distances on data,called “Dif-fusion Distance”. This definition could be applied on KernelK-means. Because thecomplexity of diffusion maps is relatively high,directly application is abandoned.This article is mainly divided into two parts: a. How to apply DiffusionDistanceto Kernel K-means algorithm and how to get a better result without consuming a largeamounts of time. b. How to get these parametres,like diffusion coefficient, in kerneltransformation by using experiment results. So that we can construct the final algo-rithm.