Fraud detection on international direct dial call using hybrid NBTree algorithm and Kullback Leibler divergence

Abstract

Fraud detection is a serious challenge in the telecommunication sector, including in international direct dial (IDD) call service. Fraudulent activities cause a greater impact on the company because the loss of revenue results in the loss of the gain due to expenses to be paid to global partners to provide an international call interconnection. Therefore IDD call fraud continues to be a concern among all IDD call service providers by developing various methods to overcome the problem. The objective of this paper is to propose a method to detect fraud suspects on IDD call services which combines the advantages of hybrid NBTree and Kullback Leibler divergence (KL-divergence or KLD). NBTree is employed due to its ability to handle large size data and its performance in accuracy and tree size that outperforms Decision Tree and Naive Bayesian. In addition, the use of KL-divergence in fraud detection, similarity measurement, and feature selection has long been proven and implemented practice. The experiment results show that the combination of the two provides better accuracy and F1-measure compared with the previous method: Naive Bayesian Classifier, hybrid Naive Bayesian — KL-divergence, and Support Vector Machine (SVM).