|
Activity Number:
|
190
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2008 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #302293 |
|
Title:
|
Identifying Statistically Significant Clusters
|
|
Author(s):
|
Ranjan Maitra*+ and Volodymyr Melnykov
|
|
Companies:
|
Iowa State University and Iowa State University
|
|
Address:
|
Department of Statistics, Ames, IA, 50011,
|
|
Keywords:
|
EM algorithm ; p-value ; second-order delta method
|
|
Abstract:
|
A formal hypothesis-testing approach is presented for detecting significant clusters in data. The approach revolves on calculating the p-value of forming an additional cluster at each stage. We illustrate our methodology using model-based clustering as an example. In this setting, we can use the second-order delta method and obtain exact p-values for the change in optimized likelihood with each additional cluster. Experiments on simulation and classification datasets are promising.
|
