基于TAN结构的启发式贝叶斯网络结构学习算法

贝叶斯网络结构学习是个NP难题。一种有效且准确性较高的学习算法是K2算法。但K2算法要确定结点次序,在无先验信息时受到很大限制。提出了一种启发式结构学习G算法,该算法以学习树扩展朴素贝叶斯TAN结构作为启发式信息,由该启发式信息生成结点次序,再用K2算法生成贝叶斯网络结构。实验结果表明,G算法可以解决无先验信息时确定结点次序的问题。所添加的弧比较简洁,网络结构比TAN结构更加合理。

BN Structure Learning Heuristic Algorithm Based on TAN Structure

The structure learning for Bayesian netwoks is NP-hard problem,K2 is one of efficacious and accurate algorithms.K2 confirms the order of nodes firstly.To a certain extent this limits in non-information.This paper purposes a new heuristic Bayesian networks structure learning G algorithm.This algorithm uses TAN structure which learns as heuristic information,using K2 algorithm learning Bayesian netwoks structure.The experimental result shows that G algorithm can solve nodes order in non-information.Arcs is sententious,comparing TAN structure,it's more reasonable.