Some silhouette-based graphics for clustering interpretation

Authors

  • E. Trauwaert Centrum S.T.O.O., V.U.B.
  • P. Rousseeuw Centrum S.T.O.O., V.U.B.
  • L. Kaufman Centrum S.T.O.O., V.U.B.

Abstract

Silhouettes were developed as a graphical display for nonhierarchical cluster analysis. They are based on the ratio between the tightness of a cluster and its separation from other clusters. A possible extension is to represent for each object both these characteristics in a two dimensional graph. The same technique can also be used with fuzzy clustering, making use directly of the fuzzy membership functions to measurethe tightness ofthe links of each objectwith its principal
cluster and its neighbour.

Downloads

Published

1989-09-01

How to Cite

Trauwaert, E., Rousseeuw, P., & Kaufman, L. (1989). Some silhouette-based graphics for clustering interpretation. JORBEL - Belgian Journal of Operations Research, Statistics, and Computer Science, 29(3), 35–55. Retrieved from https://orbel.be/jorbel/index.php/jorbel/article/view/86

Issue

Vol. 29 No. 3 (1989)

Section

Articles