The concept of random cluster based outlier detection

Abstract

Detection of outliers is one of the most common and important problems in modern data analysis. Sources of outliers are different. These could be the result of a database malfunction or user errors. The problem is very important due to the dynamic development of large data sets. Therefore, in this paper we present detailed results of work on the concept of using distribution properties to detect outliers. The aim of the study is to introduce an innovative solution that enables the use of statistical semantics of identification and classification of outliers. The undoubted advantages of the novel approach for outlier detection are the simplicity of interpretation and the possibility of its modification. The effectiveness of the proposed method was compared with other recognized techniques to detecting outliers on both artificially generated and empirical data sets.