Improved heuristic equivalent search algorithm based on Maximal Information Coefficient for Bayesian Network Structure Learning

Abstract

Greedy Equivalent Search (GES) is an effective algorithm for Bayesian network structure learning problem, which searches in the space of graph equivalence classes. However, original GES which takes greedy strategy into account may easily fall into local optimization trap because of the empty initial structure. In this paper, an improved GES method is proposed. It firstly designs a draft of the real network, based on conditional independence tests and Maximum Information Coefficient, which helps in finding more correct dependent relationship between variables. To ensure correctness, this draft is used as a seed structure of original GES algorithm. Numerical experiments on four standard networks show that SCo (the value of the BDeu score) and NEtoGS (the number of graph structure, which is equivalent to the Gold Standard network) have big improvement. Also, the total of learning time is greatly reduced. Therefore, our improved method can relatively quickly determine the structure with highest degree of data matching.