Interpretability-based validity methods for clustering results evaluation

摘要

Validation and interpretation are the two last steps of a clustering process. Generally these steps are processed separately since the existing validity measures are not intended to express the interpretability or the non interpretability of clusters. We propose in this paper to merge the validation and interpretation steps by using a new supervised measure that we call Homogeneity degree and which is based on the criterion of interpretability to validate clusters. We also present an extended version of this measure in order to improve its use as a relative measure.