|
Activity Number:
|
388
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 9, 2006 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #306912 |
|
Title:
|
Cluster Analysis Using Methods of Pairwise Weight on Mixed Type Attributes
|
|
Author(s):
|
William Warde*+
|
|
Companies:
|
Oklahoma State University
|
|
Address:
|
Statistics Department, Stillwater, OK, 74078-1055,
|
|
Keywords:
|
agglomerative clustering ; Rand's C statistic ; mixed-type objects ; association coefficient
|
|
Abstract:
|
A new formulation of distance is proposed for clustering mixed numeric and multiple binary data. The level of within cluster resemblance as quantified by the similarity, that is pairwise weight between i-th and j-th objects, must be accounted for in cluster analysis. The performance of clustering algorithms with our proposed distance based on methods of pairwise weight on mixed numeric and multiple binary data gives competitive or superior recovery and agreement level to Gower's. The agreement is used to quantify consistency of resultant clusterings produced by cluster algorithms. It might be a natural basis for organizing objects measured on mixed-type attributes by examining resultant clusterings depending on the characteristics of the data. The result of clustering algorithms with proposed distance based on methods of pairwise weights is examined by using principal coordinates.
|
