Constraint-based measure for estimating overlap in clustering

Different clustering algorithms have different strengths and weaknesses. Given a dataset and a clustering task, it is up to the user to choose the most suitable clustering algorithm. In this paper, we study to what extent this choice can be supported by a measure of overlap among clusters. We propose a concrete, efficiently computable constraint-based measure. We show that the measure is indeed informative: on the basis of this measure alone, one can make better decisions about which clustering algorithm to use. However, when combined with other features of the input dataset, such as dimensionality, it seems that the proposed measure does not provide useful additional information.

Publication year:2017

Accessibility:Open