基于预测能力的贝叶斯网络结构学习  

Learning Bayesian networks structure based on prediction ability

作　　者：王辉[1] 张剑飞[2] 王双成[3] 

机构地区：[1]东北师范大学计算机学院,吉林长春130024 [2]齐齐哈尔大学信息科学与电气工程学院,黑龙江齐齐哈尔161006 [3]吉林大学计算机科学与技术学院,吉林长春130025

出　　处：《东北师大学报（自然科学版）》2005年第1期32-35,共4页Journal of Northeast Normal University(Natural Science Edition)

基　　金：吉林省自然科学基金资助项目(20030517-1)

摘　　要：　给出了变量之间预测能力的概念及估计方法,证明了预测能力就是预测正确率.在此基础上建立了基于预测能力的贝叶斯网络结构学习方法,并使用模拟数据进行了实验.实验结果显示该算法能够有效地进行贝叶斯网络结构学习.In the paper,the concept and estimating methods of the prediction ability are given.At the same time,it is proven that the prediction ability is the accurate of the prediction.The method of Bayesian networks structure learning based on prediction ability is developed.This method is made up of two parts:(1)setting up elementary Bayesian networks structure according to the absolute prediction ability;(2)regulating elementary Bayesian networks structure according to the conditional prediction ability and checking the loop.The experiment is made by simulation and the results are shown in the way of contrast,experimental results show that Bayesian networks structure can be learned by this method effectively.

关 键 词：贝叶斯网络 结构学习 预测能力 条件独立 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]