An incremental Bayesian network structure learning algorithm based on local updating strategy

Abstract

Bayesian network is wildly applied in AI field, but current structure learning algorithms consume a significant amount of time when faced with big data. We propose an incremental algorithm based on local updating strategy. First, we adopted a balanced decomposition for the structure, found the sub-network needed to be updated, and updated it to obtain the new network. Then we did simulation experiments to verify the validity of our method. Finally, we compared the efficiency and accuracy of our method with that of other mainstream algorithms. The result shows that our method runs in a low constant time, and its average accuracy of structure learning reaches 98%.