The PDG-Mixture Model for Clustering

Within data mining, clustering can be considered the most important unsupervised learning problem which deals with finding a structure in a collection of unlabeled data. Generally, clustering refers to the process of organizing objects into groups whose members are similar. Among clustering approaches, those methods based on probabilistic models have been extensively developed, such as Näıve Bayes (NB) with a latent class (cluster identifier) found via an EM algorithm. Probabilistic Decision Graphs (PDGs) are a class of graphical models that can naturally encode some context specific independencies that cannot always be efficiently captured by other commonly used models. In this paper we propose to use a mixture of PDG models in cluster discovery, and an algorithm for automatic induction of the mixture and the models is introduced. The proposed approach was experimentally evaluated on both synthetic and real-world databases, and the presentation of the results includes a comparison with related techniques. The comparison demonstrates competitive performance of the mixture of PDG models with respect to likelihood. Also, the mixture of PDG models have a tendency to use fewer models (clusters) to represent domains where other models use large amounts of clusters.