Improving Bayesian network structure learning with optimized node ordering

Abstract

Learning a General Bayesian Network (GBN) depends on node ordering and correlation of nodes. The current methods are not efficient enough because of the random node ordering and not considering about the degree of independence between nodes. In this paper, we propose a new method for structure learning. It introduces the degree of independence between nodes into the scoring function, and uses genetic algorithm to search for the best ordering. With the improved scoring function and optimal ordering, the Bayesian Network for the Alert System of Industrial Boiler Security is structurally learned, and errors are compared. The experiment results show that our method is more efficient. © 2011 Springer-Verlag.