Cluster analysis has been a popular method for statistical classification. The classical cluster analysis, however, has a theoretical shortcoming in the sense that the inference to determine the number of clusters does not provide the theoretical guideline. To estimate the number of clusters, this paper explores the problem through the EM algorithm, Maximum a Posteriori and Gibbs sampler. In addition, we investigate the Bayesian Information criteria (BIC), the Laplace Metropolis criteria and the modified Fisher's criteria in order to determine the number of clusters.
Bibliographical noteFunding Information:
This research was supported by Fund for Supporting Basic Science Research in the College of Business and Economics, Yonsei University.
All Science Journal Classification (ASJC) codes
- Statistics and Probability