关键词: 贝叶斯网络, 结构学习, 互信息, 条件独立性测试, 贪婪搜索

Abstract: Structure learning is one of the most important branches in Bayesian network, while learning Bayesian network structures from data is NP-complete. An improved algorithm is provided for learning Bayesian network structures from data. It constructs the initial undirected graph based on mutual information, and then orients undirected edges by using conditional independence tests, additionally a local optimal method for four-node and five-node loops is proposed to construct the initial draft about the structure, finally greedy search is performed to explore the optimal structure. Numerical experiments show that both the BIC score and structure error have some improvements, and the number of iterations and running time are greatly reduced. Therefore the structure with highest degree of data matching can be relatively faster determined by the improved algorithm.

Key words: Bayesian network, structure learning, mutual information, conditional independence test, greedy search