Abstract
Purchase PDF (1835 K)
E-mail Article
Add to my Quick Links
Cited By in Scopus (0)
Purchase PDF (267 K)
doi:10.1016/j.datak.2007.03.006 
Copyright © 2007 Elsevier B.V. All rights reserved.
Incremental procedures for partitioning highly intermixed multi-class datasets into hyper-spherical and hyper-ellipsoidal clusters
Received 12 November 2006;
References and further reading may be available for this article. To view references and further reading you must purchase this article.
Abstract
Two procedures for partitioning large collections of highly intermixed datasets of different classes into a number of hyper-spherical or hyper-ellipsoidal clusters are presented. The incremental procedures are to generate a minimum numbers of hyper-spherical or hyper-ellipsoidal clusters with each cluster containing a maximum number of data points of the same class. The procedures extend the move-to-front algorithms originally designed for construction of minimum sized enclosing balls or ellipsoids for dataset of a single class. The resulting clusters of the dataset can be used for data modeling, outlier detection, discrimination analysis, and knowledge discovery.
Keywords: Data models; Data clustering; Mini-max partition; Geometrical approximation; Knowledge discovery
Article Outline
- 1. Introduction
- 2. Overview of geometrical data clustering
- 3. Minimum enclosing hyper-balls and hyper-ellipsoids constructions
- 4. Incremental construction of hyper-spheres and hyper-ellipsoids for multi-class data clustering
- 4.1. Move-to-front multi-hyper-ball-clustering algorithm (MFMHBC)
- 4.2. Move-to-front multi-hyper-ellipsoid-clustering algorithm (MFMHEC)
- 5. Experiment results
- 5.1. Case one moderately intermixed data set of two classes in two-dimensional space
- 5.2. Case two – three highly intermixed classes in two-dimensional space
- 5.3. Case three – Intermixed data set of three classes in three-dimensional space
- 5.4. Comparison of MFMHBC with MFMHEC in multiple dimensions
- 6. Summary and conclusions
- References
- Vitae
