Abstract:

As the decomposition of Bayesian networks (BN) into their maximal prime sub-graph is a NP-hard problem, a new method for constructing the maximal prime sub-graph junction tree (MPD-JT) of  BN is presented. This approach obtains the MPD of BN motivated by the full conditional independence information encoded in the moral graph of the original BN and uses these MPDs as a cluster of nodes to construct a junction tree, it also avoids the process of triangulation and reduces the time complexity by eliminating some coincident nodes. The correctness of the method is proven and the validity of the algorithm is analyzed by an example．