Cluster Validity by Concurrent Chaining

Abstract

A cluster is a maximal collection of suitably similar objects drawn from a larger collection of objects. Many cluster analysis procedures seek to determine a partition of objects into clusters, or more generally, to determine a hierarchy of such cluster partitions. Single-, complete-, and average-linkage are several of the most frequently cited methods that identify such a hierarchy of clusters. These simply applied linkage based procedures yield only rudimentary and often misleading information regarding the “validity” of each cluster identified. By the validity of a particular cluster we refer to the substantive significance that should be afforded that collection of objects as representing a single cluster.