|
Activity Number:
|
607
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 6, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #305188 |
|
Title:
|
Optimal Nearest Shrunken Centroids Method for High-Dimensional Classification
|
|
Author(s):
|
Tiejun Tong*+ and Herbert Pang
|
|
Companies:
|
University of Colorado and Duke University
|
|
Address:
|
, Boulder, CO, 80309,
|
|
Keywords:
|
classification ; high-dimensional data ; nearest shrunken centroids ; cross-validation ; risk function
|
|
Abstract:
|
Class prediction with the nearest shrunken centroids (NSC) method has been shown to be very successful in many high-dimensional classification problems. Nevertheless, it has two major limitations: 1) the performance of the NSC method is not satisfactory when the sample size is relatively small owing to the large variation in feature selection; 2) the NSC method is ad hoc as the tuning parameter is chosen by cross-validation. In this article, we propose a new algorithm that chooses the tuning parameter by minimizing some certain risk function. Simulation studies indicate that the proposed algorithm performs remarkably well compared to the originally NSC method by cross-validation when the sample size is small. Theoretical aspects of the tuning parameter estimation are also investigated. Finally, we conduct a real data study to evaluate the proposed findings.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2009 program |
