The Construction of Hierarchic and Non-Hierarchic Classifications

Many of the cluster methods that are used in the construction of classificatory systems operate on data in the form of a dissimilarity coefficient on a set of objects. In this paper we outline a theoretical framework within which the properties of such methods may be discussed. Certain conditions that a cluster method should satisfy are suggested, and a particular sequence of cluster methods which satisfies these conditions is described. The application of the sequence of methods is illustrated by a simple example.