基于结点排序的贝叶斯网络结构学习  

Learning Bayesian Network Structure Based on Nodes Ordering

作　　者：王双成[1] 

机构地区：[1]上海立信会计学院信息科学系,上海201600

出　　处：《计算机工程与应用》2005年第18期11-12,187,共3页Computer Engineering and Applications

基　　金：国家自然科学基金资助项目(编号:60275026);吉林省自然科学基金项目(编号:20030517-1)

摘　　要：给出了变量之间k阶分类能力的概念及计算方法,并证明了k阶分类能力就是k阶分类正确率,以及k阶分类能力和条件独立性的等价性,在此基础上构造出基于分类能力的贝叶斯网络结构打分函数,同时结合依赖分析方法和打分-搜索方法建立了有效的贝叶斯网络结构学习方法,实验结果显示该方法能够有效地进行贝叶斯网络结构学习,并使学习得到的结构倾向于简单化。The concept and algorithm of k order classification ability is presented.It is proved that the k order classification ability is the k order classification accuracy and it is equivalent with conditional independency.The scoring function of Bayesian network structure is developed.And a new method of learning Bayesian network structure is set by combining dependency analysis and search & scoring.Experimental results show that this method can effectively learn Bayesian network structure from data and learned structure is inclined to simplification.

关 键 词：贝叶斯网络 结构学习 分类能力 打分函数 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]