An overview of outlier detection methods

Abstract

Outlier detection is a necessary task; in many safety critical environments such as aircraft engine rotation failure, factory production line, network intrusion, and bust analysis. Most proposed methods are based on distance and density based outlier detection which simply detects outliers by calculating distances or density between the data points. In high dimensional datasets, it's very difficult to find outliers with the measure of distance or density methods, because in such spaces the data become sparse and the imagination of data distribution is also hard. So with the curse of dimensionality we discuss a cluster based method that examines the behavior of the data in low dimensional projection. On the other hand most existing methods detect outliers with some parameters that needed to be defined by users in advance. In some cases it is very difficult for users to define these parameters. To solve such problems, example based method has been proposed. In this paper, wide and different modem technologies of outlier detection will be considered.