An Empirical Study of Efficiency and Accuracy of Probabilistic Graphical Models

In this paper we compare Na\ii ve Bayes (NB) models, general Bayes Net (BN) models and Probabilistic Decision Graph (PDG) models w.r.t. accuracy and efficiency. As the basis for our analysis we use graphs of size vs. likelihood that show the theoretical capabilities of the models. We also measure accuracy and efficiency empirically by running exact inference algorithms on randomly generated queries. Our analysis supports previous results by showing good accuracy for NB models compared to both BN and PDG models. However, our results also show that the advantage of the low complexity inference provided by NB models is not as significant as assessed in a previous study