基于聚类的离群点分析方法

Analysis algorithm for cluster-based outlier

对于离群点的形成,不同的属性起着不同的作用,离群点在不同的属性域中,会表现出不同的离群特性,在大多数情况下,高维数据空间中的对象是否离群往往取决于这些对象在低维空间中的投影。针对如何将离群点按照形成原因分类的问题,引入离群属性和离群簇等概念,以现有离群挖掘技术为基础,提出了基于离群分类来进行离群点分析的方法,实现了基于聚类的离群点分类算法CBOC(cluster-based outlier classification),以揭示离群点的内涵知识。实验表明了该方法在实际应用中的有效性。

For the formation of outliers,different attributes play different roles.Outliers in the different attribute space will show the different characteristics.In most cases,whether the high dimensional data is the outlier usually depends on the projection of these objects in low-dimensional space.In order to classify the origin of outliers,this paper defined some concepts such as outlier attributes and outlier cluster and proposed the approaches to outlier analysis by classified outliers based on the existing outlier mining technology.Furthermore,it gave an effective CBOC algorithm to provide outliers＇intensional knowledge,whose validity in practical applications is finally verified by the experimental results.