|
Activity Number:
|
467
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
ENAR
|
| Abstract - #303735 |
|
Title:
|
Lower Confidence Bounds for Prediction Accuracy in High Dimensions with Feature Selection
|
|
Author(s):
|
Kevin K. Dobbin*+
|
|
Companies:
|
The University of Georgia
|
|
Address:
|
Paul D. Coverdell Center, Room 150E, Athens, GA, 30602,
|
|
Keywords:
|
classification ; prediction ; confidence interval ; high dimensional data
|
|
Abstract:
|
Resampling-based methods work very well for obtaining a point estimate of prediction accuracy that is nearly unbiased. But constructing a corresponding confidence bound is a more difficult problem. The difficulty is both conceptual, due to the fact that prediction accuracy is not a population parameter, and computational, due to the cost of complex resampling in high dimensions. We present an alternative approach that is conceptually simpler and computationally less expensive. It extends a method developed by McLachlan for the low dimensional setting to high dimensions and high dimensions with feature selection. Simulations are used to assess nominal coverage of various methods, and applications to microarray data sets are examined.
|
- 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 |
