|
Activity Number:
|
110
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 7, 2006 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistics and the Environment
|
| Abstract - #306032 |
|
Title:
|
Testing Outliers Using a Mixture Population When Some Data Are Missing and Training Data Are Unlabeled
|
|
Author(s):
|
Aruna Saram*+ and Ferry Butar Butar
|
|
Companies:
|
Sam Houston State University and Sam Houston State University
|
|
Address:
|
2501 Lake Road, Huntsville, TX, 77340,
|
|
Keywords:
|
missing data ; EM algorithm ; outlier testing ; mixture population
|
|
Abstract:
|
This paper is concerned with the problem of multivariate outlier testing for purpose of distinguishing seismic signals of underground nuclear events from training samples based on non-nuclear seismic events when some of the data are missing and unlabeled. Suppose some of the observations are missing and we assume that training data follow a multivariate normal distribution. Using generalized likelihood ratio test procedure to perform the outlier testing and Hotelling's T2 distribution for critical values always perfect when the data are not missing. We describe an EM algorithm base procedure for using the modified likelihood ratio test to test for outliers when the training data follow a mixture distribution and some of the observations are missing. We use seismic data and simulated data to describe this procedure.
|
