基于MH算法的贝叶斯网络结构学习的改进  

Improvement of Bayesian Network Structure Learning Based on MH Algorithm

作　　者：胡雨 陈雪东[2] HU Yu;CHEN Xuedong(Faculty of Mathematical and Information,Zhejiang Normal University,Jinhua 321000,China;School of Science,Huzhou University,Huzhou 313000,China)

机构地区：[1]浙江师范大学数理信息与工程学院,浙江金华321004 [2]湖州师范学院理学院,浙江湖州313000

出　　处：《湖州师范学院学报》2018年第2期6-11,共6页Journal of Huzhou University

基　　金：国家自然科学基金项目(11171105)

摘　　要：针对贝叶斯网络结构学习中的MCMC(马尔科夫链蒙特卡洛)算法,提出一种新的建议分布(概率)以改进其中的Metropolis-Hastings(MH)算法,使其相应的MCMC算法提升了贝叶斯网络结构学习的收敛速度.同时,将新的算法应用于携程网房型产品的用户行为数据集,即一个有监督的不平衡分类问题,最后将分析结果与随机森林、logistic回归方法相对比.实验结果表明,新的MCMC实现图模型具有较好的预测效果.For the MCMC(Markov Chain Monte Carlo)algorithm in Bayesian network structure learning,this paper proposes a new proposal distribution(probability)of the Metropolis Hastings(MH)algorithm to improve the convergence rate of the corresponding Bayesian network structure learning.At the same time,the proposed algorithm is applied to the user behavior data set of Ctrip’s room-type products,that is,a supervised imbalanced classification problem.Finally the analysis results are compared with the random forest and logistical regression methods.Experimental results show that the new MCMC realization graph model has a good prediction effect.

关 键 词：叶斯网络 MCMC算法 建议概率 结构学习 不平衡数据 

分 类 号：O212[理学—概率论与数理统计]