基于MARS和概率规划的离群值检测算法  

Outlier detection algorithm based on MARS and probabilistic programming

作　　者：王瑞豪 童英华[1,2] 冯忠岭[3] WANG Rui-hao;TONG Ying-hua;FENG Zhong-ling(School of Computer,Qinghai Normal University,Xining 810008,China;The State Key Laboratory of Tibetan Intelligent Information Processing and Application,Qinghai Normal University,Xining 810008,China;School of Physics and Electronic Information,Qinghai Normal University,Xining 810008,China)

机构地区：[1]青海师范大学计算机学院,青海西宁810008 [2]青海师范大学省部共建藏语智能信息处理及应用国家重点实验室,青海西宁810008 [3]青海师范大学物理与电子信息学院,青海西宁810008

出　　处：《计算机工程与设计》2023年第9期2694-2699,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(61862055);河北省物联网监控工程技术研究中心基金项目(3142016020);青海省物联网重点实验室基金项目(2020-ZJ-Y16)。

摘　　要：为提高物联网数据质量,提出一种基于MARS模型和概率规划的多变量离群值检测方法。该方法能够将多个变量结合起来,通过一个模型检测离群值。创建一个多元自适应回归样条模型产生研究化的残差,将残差作为输入,采用概率规划方法,建立基于全贝叶斯推理的一般单变量离群点检测模型。实验结果表明,概率规划模型能检测出更准确的离群点,该模型提供了具有可信区间的概率分布。To improve the data quality of internet of things,a multi-variable outlier detection method based on multivariate adaptive regression spline(MARS)model and probability programming was proposed.This method can combine multiple variables and detect outliers through a model.A multivariate adaptive regression spline model was built to generate the research residuals.A general univariate outlier detection model based on full Bayesian inference was established by using probabilistic programming method with residual as input.The results show that the probabilistic programming model can detect more accurate outlier points,and the model provides a probability distribution with confidence interval.

关 键 词：物联网 数据质量 多元自适应回归模型 离群值 概率规划 多元离群点检测 贝叶斯 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]