Hits:

Indexed by:期刊论文

Date of Publication:2011-01-01

Journal:ICIC Express Letters

Included Journals:EI、Scopus

Volume:5

Issue:1

Page Number:155-161

ISSN No.:1881803X

Abstract:In order to evaluate the performance and estimate the cluster number in the spectral clustering, a cluster validation criterion called PerfectRatio based on multiway normalized cut criterion is defined. PerfectRatio measures the degree of perfect of the similarity matrix through the ratio of the total normalized association of the original data set to the sum of the leading eigenvalues of corresponding stochastic transition matrix. One of the good features of this measure is that it can reflect the hierarchy of a data set under a properly chosen Gaussian kernel parameter of the similarity matrix. Numerical experiments are illustrated to demonstrate the effectiveness of the proposed method. ? 2011 ISSN.