|
Activity Number:
|
289
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 8, 2006 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #306616 |
|
Title:
|
Clustering of Outlier Structure Using Minimal Spanning Tree Rankings and Minimum Volume Ellipsoids
|
|
Author(s):
|
Mark W. Lukens*+ and James Gentle
|
|
Companies:
|
George Mason University and George Mason University
|
|
Address:
|
4061 Cardinal Crest Drive, Woodbridge, VA, 22193,
|
|
Keywords:
|
clustering ; outliers ; minimal spanning tree ; minimum volume ellipsoids ; multivariate ranking
|
|
Abstract:
|
One method to explore high dimensional data structure is through the use of minimal spanning trees. Multivariate rankings might be used in conjunction with a minimal spanning tree as a way of ordering the data. The edge distance between rankings is one way to separate the tree into clusters. This technique, however, depends upon how the ranking is done. Use of rankings from minimal spanning trees might encounter difficulties when multiple groups appear in the data. A different approach is to use minimum volume ellipsoids and a minimal spanning tree to rank the multivariate data. The minimum volume ellipsoids define the main structure and then the minimal spanning tree is fitted to the remainder of the outlier data. This not only provides a boundary between the main structure and any outlier groupings but can also assist in the identification of clusters.
|
