贝叶斯网络参数学习中的连续变量离散化方法研究  

Research on Discretization Methods of Continuous Variables of Parameter Learning in Bayesian Network

作　　者：刘晓明[1] 李盼池[1] 刘显德[1] 肖红[1] LIU Xiaoming;LI Panchi;LIU Xiande;XIAO Hong(School of Computer and Information Technology,Northeast Petroleum University,Daqing 163318)

机构地区：[1]东北石油大学计算机与信息技术学院,大庆163318

出　　处：《计算机与数字工程》2018年第5期992-996,共5页Computer & Digital Engineering

摘　　要：在贝叶斯网络参数学习中,通常假定所有变量是离散变量或者服从高斯分布的连续变量,所以需要对现实中一些不符合该前提假设的变量进行离散化。连续变量的离散化直接影响贝叶斯网络的推理效果,因而具有重要意义。论文采用两种不同的方法(分别为等宽法、ChiMerge法)对数据集进行离散化,离散后的结果运用Netica构建相应的贝叶斯网络并进行参数学习,最后利用得到的贝叶斯网络进行简单的预测分析。It can qualitatively and quantitatively analyze the dependencies between attributes,and do probabilistic reasoning.In parameter learning of Bayesian network,it is usually assumed that all variables are discrete or continuous variables obeying Gaussian distribution,so it is necessary to do the discretization for those variables in reality which disobey the assumptions. In this paper,two different methods（equal width,ChiMerge）are used to discrete data set,then the results of discretization are used to construct the corresponding Bayesian network and do parameter learning in Netica. Finally,the resulting Bayesian network is used to do some simple forecast and analysis.

关 键 词：贝叶斯网络 连续变量 离散化 

分 类 号：TJ765.2[兵器科学与技术—武器系统与运用工程]