|
Activity Number:
|
433
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #305029 |
|
Title:
|
Penalized Model-Based Clustering with General Covariance Matrices
|
|
Author(s):
|
Hui Zhou*+ and Wei Pan and Xiaotong Shen
|
|
Companies:
|
The University of Minnesota and The University of Minnesota and The University of Minnesota
|
|
Address:
|
A460 Mayo Building, MMC 303, , Minneapolis, MN, 55455,
|
|
Keywords:
|
cluster ; EM ; Penalized Likelihood ; shrinkage ; glasso ; Microarray
|
|
Abstract:
|
Clustering analysis is one of the most important methods in analyzing microarray and other high-dimensional data, yet the presence of many noise variables may mask underlying clustering structures. Hence simultaneous parameter estimation and removal of noise variables via variable selection are essential. Existing penalized likelihood approaches in model-based clustering analysis all assume diagonal covariance matrices, which however may not hold in practice. Here we consider an approach that could incorporate cluster-specific general covariance matrices. We derive an EM algorithm with the use of the graphical lasso for parameter estimation. Simulation studies and applications to microarray gene expression data demonstrate the utility of the proposed method.
|
