Structural learning of Bayesian networks from complete data using the scatter search documents

Bayesian networks are directed acyclic graphs that model the dependency relationships between variables of interest. These networks are characterized by the structure of the network and the conditional probabilities that specify the dependencies that exist between the variables. In this paper, the scatter search optimization algorithm is utilized in learning the structure of the Bayesian network from complete data. This involves a heuristic search for the best network structure that maximizes a scoring function given a database of cases. The scatter search algorithm is implemented on a database of cases sampled from known networks in order to test the accuracy of the structural learning. Empirical results from the implementations are presented in the paper.