|
Activity Number:
|
401
|
|
Type:
|
Invited
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
IMS
|
| Abstract - #302816 |
|
Title:
|
A Prediction Interval for the Misclassification Rate
|
|
Author(s):
|
Susan Murphy*+ and Eric Laber
|
|
Companies:
|
University of Michigan and University of Michigan
|
|
Address:
|
Dept. of Statistics, Ann Arbor, MI, 48109-1107,
|
|
Keywords:
|
classification ; confidence sets
|
|
Abstract:
|
The misclassification rate of a classifier is a non-smooth functions of the classifier. The estimated rate suffers from bias due to over-fitting and the misclassification rate is a "minimized" quantity. For both of these reasons the construction of measures of confidence such as estimates of variance and confidence/prediction intervals are challenging. We discuss this problem and propose a method based on the use of a smooth upper bound combined with the bootstrap. This upper bound utilizes the surrogate loss that is used in the construction of the classifier.
|
