|
Activity Number:
|
343
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, July 31, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #308819 |
|
Title:
|
Stalactite Plot for Outlier Detection in the Presence of a Computationally Singular Covariance Matrix
|
|
Author(s):
|
Jeremy Nadolski*+ and Pablo Marquez and Lee Ann Smith
|
|
Companies:
|
Benedictine University and Benedictine University and Benedictine University
|
|
Address:
|
5700 College Rd, Lisle, IL, 60532,
|
|
Keywords:
|
outlier detection ; stalactite algorithm ; singular value decomposition ; Mahalanobis distance
|
|
Abstract:
|
Outlier detection is critical in any data analysis project. Outliers affect the ability to predict outcomes successfully and deteriorate accuracy. We used an improved stalactite algorithm proposed by Szychowski et al (2005) to detect outliers in our data. The algorithm is reliant upon Mahalanobis distance which uses the inverse of the covariance matrix. As such, non-invertible matrices in our datasets, which were either noninvertible or computationally non-invertible, could not be supported. To circumvent the non-invertible matrix, we used Singular Value Decomposition (SVD) to create a pseudo-inverse. Using the pseudo-inverse, we were able to replicate the results of Szychowski et al. (2005) and complete our algorithm to find outliers. Our outlier detection algorithm is applied to a dataset on Drosophila melanogaster egg-laying behavior from different wild type strains.
|
