Intelligent control of the hierarchical agglomerative clustering process

The basic process of Hierarchical Agglomerative (HAG) clustering is described as a merging of clusters based on their proximity. The importance of the selected cluster distance measure in the determination of resulting clusters is pointed out. We note a fundamental distinction between the nearest neighbor cluster distance measure, Min, and the furthest neighbor measure, Max. The first favors the merging of large clusters while the later favors the merging of smaller clusters. We introduce a number of families of intercluster distance measures each of which can be parameterized along a scale characterizing their preference for merging larger or smaller clusters. We then consider the use of this distinction between distance measures as a way of controlling the hierarchical clustering process. Combining this with the ability of fuzzy systems modeling to formalize linguistic specifications, we see the emergence of a tool to add human like intelligence to the clustering process.