文献类型：会议论文

英文题名：Study on projected outlier detecting algorithm in high dimensional space

作者：Chen, JB; Gao, YM

第一作者：陈建斌

通讯作者：Chen, JB[1]

机构：[1]Beijing Union Univ, Dept eBusiness, Coll Business, Beijing 100025, Peoples R China

第一机构：北京联合大学商务学院

通讯机构：[1]corresponding author), Beijing Union Univ, Dept eBusiness, Coll Business, Beijing 100025, Peoples R China.|[1141721]北京联合大学商务学院;[11417]北京联合大学;

会议论文集：6th International Symposium on Test and Measurement (ISTM)

会议日期：JUN 01-04, 2005

会议地点：Dalian, PEOPLES R CHINA

语种：英文

外文关键词：outlier detecting; projected clustering; attribute entropy

摘要：Based on the discussion of the outlier detecting methods in high dimensional data space, a new projected outlier detecting algorithm has been proposed in this paper Not as the traditional algorithms that find outliers in full dimension space, our algorithm is extended from projected clustering techniques, and turn to the subspace to detect outliers. Firstly, the relationship of the outliers(little pattern) and the clusters(big pattern) has been discoursed upon. Secondly, clusters with associated subset of attributes have been identified. By constructing an entropy metric based on similarity between objects, the Decentralization of each attribute can be evaluated and a subset in which attributes are more decentralization can be obtained. The outliers can be divided into two kinds. The first kind is that they are outlandish along single attributes, and the second kind is that they are outlandish in multi attributes. The algorithm can find and list them respectively. The complexity analysis has shown that the proposed algorithm is high efficiency.