摘要
针对目前的动态贝叶斯网络主要用于时间序列的因果分析和分类预测,缺少将动态贝叶斯网络用于回归计算方面研究的情况,结合随机树生成、回归变量的离散化、类变量的数量化、类的满条件概率计算和加权平均回归计算等建立动态随机树贝叶斯回归模型,并通过集成(平均)来提高回归模型的泛化能力,使用期货数据进行实验,实验结果显示,动态随机树贝叶斯集成回归模型具有良好的回归可靠性.
At present,the dynamic Bayesian network is mainly used for causality analysis and classification prediction of time series and the research on the dynamic Bayesian networks in regression computing is needed. In this paper,we combine the generation of random tree,the discretization of regression variable,the quantification of class variable,the full conditional probability calculation of a class,weighted mean regression calculation and so on to set up a dynamic random tree Bayesian regression model and improve it’ s generalization ability by model average. We use futures data to carry out experiments. Experimental results show that the dynamic random tree Bayesian ensemble regression model has good regression reliability.
作者
王双成
郑飞
唐晓清
WANG Shuang-cheng;ZHENG Fei;TANG Xiao-qing(School of Information Management,Shanghai Lixin University of Accounting and Finance,Shanghai 201620,China;School of Statistic and Mathematics,Shanghai Lixin University of Accounting and Finance,Shanghai 201620,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2019年第4期715-720,共6页
Journal of Chinese Computer Systems
基金
国家社会科学基金项目(18BTJ020)资助
上海市自然科学基金项目(15ZR1429700)资助
关键词
动态贝叶斯网络
随机树
回归模型
模型平均
回归可靠性标准
dynamic Bayesian network
random tree
regression model
model average
regression reliability criterion
