|
Activity Number:
|
416
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #305613 |
|
Title:
|
Bootstrapping for Significance in Multidimensional Compact Clustering Models
|
|
Author(s):
|
Ranjan Maitra*+ and Soumendra N. Lahiri and Volodymyr Melnykov
|
|
Companies:
|
Iowa State University and Texas A&M University and Iowa State University
|
|
Address:
|
Department of Statistics, Ames, IA, 50011,
|
|
Keywords:
|
k-means algorithm ; signficance measures ; hierarchical clustering ; compact clustering ; nonparametric bootstrap ; clustering complexity
|
|
Abstract:
|
We develop nonparametric bootstrap methodology for assessing measures of significance in compact clustering models. Since the objective function for any well-optimized clustering model improves with increasing complexity, we test for significance of this improvement. A nonparametric bootstrap scheme is proposed for simulating from the less complex (null) model. The methodology is applicable to multi-dimensional models: indeed performance improves with increasing dimensionality, provided the objective function is well-optimized at those dimensions. Extensive studies on simulation and classification data sets done using k-means and hierarchical clustering show excellent performance and utility of the derived methodology.
|
