On the use of hierarchical clustering in fuzzy modeling

Some methods of fuzzy clustering need to use a priori knowledge about the number of fuzzy classes or some other information about the possible distribution of the clusters. A way to improve these methods is to use hierarchical clustering as a preprocessing of the data. This approach does not provide a simple partition of the data set, but a hierarchy of them. In this paper we define several measures using fuzzy-set tools, to establish a ranking between the different possible partitions. The characteristics and properties of these criteria are studied. The paper finishes with some remarks about the use of these results in different unsupervised learning situations.