作 者: ;
机构地区: 广东科贸职业学院
出 处: 《计算机仿真》 2010年第7期179-182,共4页
摘 要: 研究统计方法分析问题,针对在实际应用外特性模型的输入普遍为混合变量,既包括连续随机变量,也包括离散随机变量。目前已有混合多元回归学习模型大多只处理连续随机变量,且有着多重共线性的缺陷。针对上述问题,研究了基于贝叶斯网络的回归树学习模型。基于贝叶斯网络的回归树学习模型的研究方法建立在朴素贝叶斯网络模型基础上,采用分而治之的原则构造决策树,以朴素贝叶斯取代叶节点。在2个UCI机器学习数据集上的仿真实验结果表明模型性能良好。基于贝叶斯网络的回归树学习模型可以有效减小预测误差。 In practical use,large-scale inputs of external characteristic model are hybrid variables,including continuous random variables and discrete random variables.Most of existing combining multiple regression learning models are not able to solve the problem of multicollinearity on hybrid variables.To solve this problem,a Bayesian-network based Regression Tree Learning Model has been deduced in this paper.This improved model is based on Native Bayesian Network Model.Our model builds tree model by the divide - and - conquer method,and replaces leaf node by applying Native Bayesian Network for Regression.The experiment results on two UCI data sets show that the improved model has a good performance,which can effectively reduce the prediction error.
领 域: [自动化与计算机技术]