基于多项式回归的Pair-Copula贝叶斯网络模型  

Pair-Copula Bayesian Network Model Based on Polynomial Regression

作　　者：牛岩溪 梁冯珍 Niu Yanxi;Liang Fengzhen(School of Mathematics,Tianjin University,Tianjin 300350,China)

机构地区：[1]天津大学数学学院,天津300350

出　　处：《统计与决策》2019年第1期24-28,共5页Statistics & Decision

摘　　要：Pair-Copula贝叶斯网络模型是解决变量间相依关系推断问题的一种有效模型,而条件独立性检验是该模型构建过程中的关键步骤。文章在改进的PC算法的基础上,提出了基于多项式回归残差的条件独立性检验方法,并进行仿真模拟实验。该方法可以良好地检验变量间的条件独立关系,通过有向无环图反映网络中的相依和独立关系,并结合Pair-Copula得到完整的相依关系推断模型以及相应的密度函数。Pair-Copula Bayesian network model is an effective model to solve the problem of inferring relationship between variables, and conditional independence testing is a key step in the model construction. Based on the improved PC algorithm, this paper proposes a conditional independence test method based on polynomial regression residuals and then carries out the simulation experiment. The proposed method can be used to effectively test the conditional independent relationship between variables.A directed acyclic graph is used to reflect the dependence and independence of a network, and Pair-Copula is combined to obtain a complete dependency relationship inference model and the corresponding density function.

关 键 词：pair-Copula贝叶斯网络 条件独立性 多项式回归 

分 类 号：O212[理学—概率论与数理统计]