|
Activity Number:
|
139
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2008 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #302369 |
|
Title:
|
Fast Diagnostics
|
|
Author(s):
|
Gentiane Haesbroeck*+
|
|
Companies:
|
University of Liege
|
|
Address:
|
Grande traverse, 12, Liege, International, 4000, Belgium
|
|
Keywords:
|
Combinatorial optimization ; Outlier detection ; Rayleigh test of uniformity
|
|
Abstract:
|
A general problem arising in diagnostics is to identify subsamples of given size whose deletion causes maximal changes in a statistic of interest as measured by an appropriate target function. When the size of the subset is equal to one, diagnostics are usually easily derived. However, this 'delete-one' approach may suffer from the masking effect and should be completed with 'delete-m' diagnostics when necessary. However, due to the underlying combinatorial problem (choose m out on n observations), the required task becomes repetitive and highly time-consuming. The aim of the talk is to present an algorithmic way to speed up the process when the target function can be expressed in terms of probability vectors. Several applications are illustrated: outlier detection in multivariate data, detection of influential points for the Rayleigh test of uniformity.
|
- 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 2008 program |
