摘要
贝叶斯网络是研究不确定环境下知识表示和因果推理的有效工具之一。MMHC算法是一种较新的贝叶斯网络结构学习算法。在MMHC算法的基础上,对几种广泛使用的贝叶斯网络评分准则如MIT、K2Score、MDL、BDeu评分准则等进行了研究,实验结果表明K2评分准则在MMHC学习算法上具有最好的学习效果,MIT评分和BDeu评分次之,MDL评分效果最差。
Bayesian network is one of the important knowledge representation and reasoning tools under uncertain conditions. Tsamardinos present a new algorithm for Bayesian network structure learning, called Max-Min Hill-Climbing （MMHC）. We apply several widely used score metrics of Bayesian network such as MIT, K2 score, BDeu and MDL to MMHC algorithm and evaluate those score metrics on four data sets. Detailed results of a complete experiment show that the K2 score metric in this algorithm is the best, and the MIT and BDeu score metric secondly, the MDL score is the worst.
出处
《西南科技大学学报》
CAS
2008年第2期56-61,共6页
Journal of Southwest University of Science and Technology
基金
国家自然科学基金(No.60473032)
教育部科学技术重点项目(No.105018)
