SUBSCRIPTION FRAUD DETECTION USING AUTOENCODER IN CASE OF ETHIOTELECOM

Abstract

Nowadays, telecom services are becoming an essential communication and business facilitators. However, the development of telecom services motivates fraudsters for illegal use. Hence, telecom fraud becomes a serious challenge in telecommunication sector by losing telecom revenue. It also results in poor quality of services for their customers. Telecom data, SMS and voice call are not free from security issues. Despite SMS, and USSD services becomes good options over installable mobile based or web based applications due to lesser cost and real time support, the methods provide fixed amount requirements which is not dynamic way to address the need of these services subscribers. The other challenge of these services is that; the services are not safe from frauds as such, especially in GSM switches. Subscription fraud is one type of fraud in today’s telecom business which is a common telecom fraud. The need of fraudsters is to make money illegally or getting telecom services with the intention of not to pay for the service they used. The prime objective of this study is to build a predictive model using deep auto-encoder to detect subscription fraud in case of Ethiotelecom. The performance of fraud detection model is 98.95% validation accuracy on ethiotelecom call detail record dataset using deep neural network autoencoder and CNN-LSTM autoencder algorithms at threshold of 0.0103 and 0173 respectively.