具有超父结点时间序列贝叶斯网络集成回归模型

发布时间：2018-02-10 00:56

本文关键词： 贝叶斯网络高斯函数时间序列分类回归数据挖掘　出处：《计算机学报》2017年12期 　论文类型：期刊论文

【摘要】：时间序列是现实世界中数据的主要表现形式之一,对时间序列进行预测也有着普遍的需求.现已发展了许多时间序列(单时间序列或多时间序列)预测模型,它们各有特点,被广泛用于解决诸多领域的实际问题,但这些模型或者需要时间序列平稳性和具有线性关系的假设,或者与某种分布紧密联系在一起,这使其适用范围受到限制,而且也不易于实现动态和静态信息的融合.文中在基于高斯函数估计属性密度的基础上,结合转换数据集构建、回归变量的离散化、类变量的数量化、属性联合密度的分解计算和以类的满条件概率为权重的加权平均等,建立用于时间序列预测的具有超父结点贝叶斯网络回归模型,该模型能够在统一的概率框架下实现对动态与静态信息的融合,不需要平稳性、分布和函数形式的假设,并能够通过具有不同超父结点贝叶斯网络回归模型的集成来进一步降低回归误差和提高泛化能力.使用UCI和宏观经济数据进行实验的结果显示,无论对单时间序列还是多时间序列,具有超父结点贝叶斯网络集成回归模型均具有良好的回归可靠性.
[Abstract]:Time series is one of the main forms of data in the real world, and there is also a general demand for time series prediction. Many time series (single time series or multiple time series) prediction models have been developed, each of which has its own characteristics. Are widely used to solve practical problems in many fields, but these models either require assumptions of time series stationarity and linear relationships, or are closely associated with certain distributions, which limit their scope of application. It is also difficult to realize the fusion of dynamic and static information. Based on the estimation of attribute density based on Gao Si function, this paper combines the construction of transformation data sets, the discretization of regression variables and the quantization of class variables. A Bayesian network regression model with super-parent nodes is established for time series prediction by decomposition of attribute joint density and weighted average weighted by full conditional probability of class. The model can realize the fusion of dynamic and static information under the unified probability framework, without the assumption of stationarity, distribution and function form. It can further reduce the regression error and improve the generalization ability by integrating Bayesian network regression models with different super-parent nodes. The experimental results using UCI and macroeconomic data show that, For both single time series and multiple time series, the integrated regression model with super-parent node Bayesian network has good regression reliability.
【作者单位】：上海立信会计金融学院信息管理学院;上海立信会计金融学院统计与数学学院;
【基金】：国家自然科学基金(61272209) 上海市自然科学基金(15ZR1429700) 上海市教委科研创新项目(15ZZ099)资助
【分类号】：O212.1

【相似文献】