Structural Learning of Bayesian Networks with Mixtures of Truncated Exponentials

In this paper we introduce a hill-climbing algorithm for structural learning of Bayesian networks from databases with discrete and continuous variables. The process is based on the optimisation of a metric that measures the accuracy of a network penalised by its complexity. The result of the algorithm is a network where the conditional distribution for each variable is a mixture of truncated exponentials (MTE), so that no restrictions on the network topology are imposed. The using artificial and real world data