An improved bayesian networks learning algorithm based on independence test and MDL scoring
Junzhong Ji, Jing Yan, Chunnian Liu, Ning Zhong · 2005
In recent years, more and more people studied the Bayesian networks learning algorithm that integrates independence test with scoring metric. Based on the proposed hybrid algorithm I-B&B-MDL, a modified method is developed. There are two major contributions. Firstly, order-0 and partial order-1 independence tests are used to obtain an original graph of the network, which reduces the number of independence tests and database passes while effectively restricting the search space. Secondly, by means of the heuristic knowledge of mutual information, sort order for candidate parent nodes increases the cut-offs of the B&B search tree and accelerates search process. The experimental results show that the modified algorithm has high accuracy, and is more efficient in time complexity than other algorithms.