Supervised Classification Using Hybrid Probabilistic Decision Graphs

In this paper we analyse the use of probabilistic decision graphs in supervised classification problems. We enhance existing models with the ability of operating in hybrid domains, where discrete and continuous variables coexist. Our proposal is based in the use of mixtures of truncated basis functions. We first introduce a new type of probabilistic graphical model, namely probabilistic decision graphs with mixture of truncated basis functions distribution, and then present an initial experimental evaluation where our proposal is compared with state-of-the-art Bayesian classifiers, showing a promising behaviour.