A Hybrid Bayesian Network Structure Learning Algorithm in Equivalence Class Space

Abstract

Greedy equivalence search (GES) is a well-known Bayesian network structure learning algorithm in equivalence class space (E-space). However, the extensive search space limits the efficiency of GES. In this paper, we propose a hybrid method to improve GES. We use mutual information to determine the strongly connected components (SCCs) graph. The SCCs graph is converted to E-space, and we take it as the initial graph of GES. The experiments reveal that our proposed approach significantly prunes the search space of GES and improves the efficiency of GES. Compared with the state-of-the-art methods, our method also has excellent accuracy.