基于邻域链的数据异常点检测  

Outlier detection based on neighborhood chain

作　　者：梁绍一 韩德强[1,2] LIANG Shao-yi;HAN De-qiang(College of Electronic and Information Engineering, Xi'an Jiaotong University, Xi'an 710049, China;CETC Key Laboratory of Aerospace Information Applications, China Electronics Technology Group Corporation, Shyiazhuang 050081, China)

机构地区：[1]西安交通大学电信学院,西安710049 [2]中国电子科技集团公司航天信息应用技术重点实验室,石家庄050081

出　　处：《控制与决策》2019年第7期1433-1440,共8页Control and Decision

基　　金：国家自然科学基金项目(61573275,61671370);国家973计划项目(2013CB329405);陕西省科技计划项目(2013KJXX-46);中央高校基本科研业务费专项资金项目(xjj2016066);中国博士后科学基金项目(2016M592790);中国电子科技集团公司航天信息应用技术重点实验室高校合作课题项目(KX172600034)

摘　　要：异常点检测(outlier detection)领域的大量研究都集中于一类“基于密度的”方法,这类方法能够克服许多传统异常点检测方法的缺陷,但仍大多使用基于几何距离的方式进行数据点局部密度的估计,导致在某些情况下反直观结果的出现。针对该问题,用一种基于邻域链的方法取代传统方法进行局部密度的估计,设计新的异常点检测方法。实验结果表明,对比经典的基于密度的异常点检测方法LOF(Local outlier factor)以及几种基于LOF的改进方法,所提出的方法能够更加准确地区分正常和异常数据点,避免反直观结果的出现。Many research works in the area of outlier detection are focused on the so called "density-based" methods.Such kind of methods can counter-act many drawbacks of the traditional outlier detection methods. However, most existing density-based methods use geometric-distance-based approaches to estimate the data point’s local density, which leads to incorrect results in certain cases. To resolve the problem, the traditional local density estimation method is substituted by a neighborhood-chain-based method, and a new outlier detection method is proposed. Compared to the local outlier factor(LOF) and several of related modifications, the proposed one can find the outliers more accurately.

关 键 词：数据挖掘 异常点检测 局部密度 局部异常因子 欧氏距离 邻域链 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]