GLOF: a new approach for mining local outlier
Shengyi Jiang, Qinghua Li, Kenli Li, Hui Wang, Zhong-Lou Meng · 2004
For many data mining applications, finding the rare instances or the outliers is more interesting than finding the common patterns. In this paper, we introduce the power mean to data mining. We propose a new approach to measure the degree of an object being an outlier, which based on the nearest neighborhood and is called generalized local outlier factor (GLOF). And we propose the rule of "k /spl sigma/" (k=2 or 1.645) for outlier detection, which needn't threshold or the prior knowledge about the number of outlier in dataset. We analyzed the formal properties of GLOF. Finally, we give empirical analysis to demonstrate the effectiveness, the experimental results show that in some cases GLOF can measure the local outlier more accurately that LOF, CBLOF, RNN. The rule of "k /spl sigma/" is promising in practice.