Sankhya: The Indian Journal of Statistics

1993, Volume 55, Series A, Pt. 3, 494--505

MAHALANOBIS METRICS FOR CLUSTER ANALYSIS

By

R. GNANADESIKAN, J. W. HARVEY, Rutgers University

And

J. R. KETTENRING, Bellcore

 

SUMMARY. Standard clustering algorithms often fail to detect clear-cut homogeneous elliptical clusters. This problem can be corrected by developing an appropriate Mahalanobis type of metric. An algorithm for this, which does not require advance knowledge of the cluster structure, was proposed by Art et al. (1982). The purpose of this paper is ti refine the algorithm, develop a graphical aid for implementing it, and report on its effectiveness.

Subject classification. 62H30

FULL PAPER.

This article in Mathematical Reviews.