The use of linked line segments for cluster representation and data reduction

Regardless of how one searches for clusters in data, the resulting cluster structures are often modelled as their respective centroids. However, the u se of centroids in this manner implies the clusters are hyperspherical in s hape. If the cluster shape is not hyperspherical, using the distance betwee n individual observations and the centroid as a metric of an observation's cluster membership can be misleading. This report proposes the use of a lin ked line segment based model of cluster structure which is not biased towar ds any particular cluster shape. The effectiveness of the linked line segme nt approach is demonstrated in a data reduction exercise using simulated an d real world data. (C) 1999 Published by Elsevier Science B.V. All rights r eserved.