【作者】 魏藜； 钱卫宁； 周傲英；

【机构】 复旦大学计算机科学与工程系；

【摘要】 <正>1引言离群点发现(outlier detection)是数据挖掘(datamining)的一项重要技术,其目标是发现数据集中行为异常的少量数据对象,这在网络监控、金融欺诈、数据清洗等许多领域都有着很强的应用背景。对于不同数据区域之间密度差别较大的数据而言,基于密度的局部离群点定义比较有效;对于分布稀疏的高维数据,需要在子空间中定义离群点以克服"维更多还原

【Abstract】 Outlier detection is a procedure to find small amount of exceptional objects that deviate from the rest of the data set.This paper presents a novel subspace local outlier test algorithm SLOT.It estimates the bounds of LOF value in the lower subspaces according to the info in the upper subspaces,so as to avoid the computation-intensive distance computation.Based on the estimated bounds of LOF value,SLOT can prune large amount of unrelated objects in advance,and as the result,it reduces the computation cost dramatically.Both formal analysis and comprehensive performance evaluation show that our method is efficient to find all local outliers in all subspaces from high-dimensional categorical data.更多还原