高维空间中的离群点发现

在许多KDD(knowledge discovery in databases)应用中,如电子商务中的欺诈行为监测,例外情况或离群点的发现比常规知识的发现更有意义.现有的离群点发现大多是针对数值属性的,而且这些方法只能发现离群点,不能对其含义进行解释.提出了一种基于超图模型的离群点(outlier)定义,这一定义既体现了"局部"的概念,又能很好地解释离群点的含义.同时给出了HOT(hypergraph-based outlier test)算法,通过计算每个点的支持度、隶属度和规模偏差来检测离群点.该算法既能够处理数值属性,又能够处理类别属性.分析表明,该算法能有效地发现高维空间数据中的离群点.

Finding Outliers in High-Dimensional Space

For many KDD (knowledge discovery in databases) applications, such as fraud detection in E-commerce, it is more interesting to find the exceptional instances or the outliers than to find the common knowledge. Most existing work in outlier detection deals with data with numerical attributes. And these methods give no explanation to the outliers after finding them. In this paper, a hypergraph-based outlier definition is presented, which considers the locality of the data and can give good explanation to the outliers,and it also gives an algorithm called HOT(hypergraph-based outlier test) to find outliers by counting three measurements,the support,belongingness and deviation of size,for each vertex in the hypergraph.This algorithm can manage both numerical attributes and categorical attributes.Analysis shows that this approach can find the outliers in high-dimensionsal space effctively.