An experimental study of partition quality indices in clustering

Abstract

We present a preliminary study to define a comparison pro- tocol to evaluate different quality measures used in supervised and un- supervised clustering as objective functions. We first define an order on the set of partitions to capture the common notion of a good partition towards the knowing of the ideal one. We demonstrate the effciency of this approach by providing several experiments.