A mixture model approach for binned data clustering

Abstract

In some particular data analysis problems, available data takes the form of an histogram. Such data are also called binned data. This paper addresses the problem of clustering binned data using mixture models. A specific EM algorithm has been proposed by Cadez et al. ([2]) to deal with these data. This algorithm has the disadvantage of being computationally expensive. In this paper, a classification version of this algorithm is proposed, which is much faster. The two approaches are compared using simulated data. The simulation results show that both algorithms generate comparable solutions in terms of resulting partition if the histogram is accurate enough. © Springer-Verlag Berlin Heidelberg 2003.