Outlier analysis for telecom fraud detection

Abstract

Every year, the number of telecommunication fraud cases increases dramatically, and companies providing such services lose billions of euros worldwide. It has been receiving more and more attention lately mobile virtual network operators (MVNOs) which operate on top of existing cellular infrastructures of the basic operators, and at the same time are able to offer cheaper call plans. This paper is aimed to identify suspicious customers with unusual behaviour, typical to potential fraudsters in MVNO. In this study, different univariate outlier detection methods are applied. Univariate outliers are obtained using call detail records (CDR) and payments records information which is aggregated by users. A special emphasis in this paper is put on the metrics designed for outlier detection in the context of suspicious customer labelling which may support the fraud experts in evaluating customers and revealing fraud. In this research, we identified specific attributes that could be applied for fraud detection. Threshold values were found for the attributes examined, which could be used to compile lists of suspicious users.