ABSTRACT
This chapter is devoted to models and algorithms that produce hierarchies of crisp clusters from square relational (usually dissimilarity) data. These algorithms are unquestionably the most heavily used group of relational clustering methods. They can be implemented in either agglomerative (clumping) or divisive (splitting) forms. Sequential, Agglomerative, Hierarchical, and Non-overlapping (SAHN) with the CL distance behaves in a manner which is, in some sense, diametrically opposite to the behavior of SL. A dendrogram is a graphical depiction of the results of any hierarchical clustering that portrays the sequential evolution and arrangement of the clusters. This method of visually representing hierarchical clustering is very appealing and is widely seen in earlier studies that used SAHN clustering, especially in numerical taxonomy of biological data. Crowding and occlusion prevent useful visualization of hierarchical cluster structure in the input data even when the number of objects seems quite small.
