Cutting the dendrogram through permutation tests.

This paper introduces an innovative approach for detecting a sub-optimal partition starting from the dendrogram produced by a hierarchical clustering techinique. The approach exploits permutation tests and it can be used regardless of the agglomeration method and distance measure used in the classification process because it relies on the same criteria used for producing it. Moreover, the proposed approach can detect partitions not necessarily identifiable using a traditional cut approach, as the resulting clusters could correspond to different heights of the tree.