SUBSCRIPTION FRAUD DETECTION USING CONVOLUTIONAL AUTO ENCODER AND DNN APPROACH IN CASE OF ETHIO TELECOM

Abstract

Nowadays, telecom services are becoming essential communication and business facilitators. However, the development of telecom services motivates fraudsters for illegal use. Hence, Telecom fraud becomes a serious challenge in the telecommunication sector which leads telecom companies to lose yearly profits and to deliver unfortunate quality of services for their Subscribers. Subscription fraud is a common and significant type of telecom fraud in today's business. The primary goal of the fraudsters is to make money illegally or to obtain telecom services with the intent of not paying for the service they used. Ethiotelecom is one of the oldest service providers in Africa and the sole service provider of Ethiopia which offers telecommunication services and products for the enhancement and development of the nation. In this paper, we use a predictive model using deep learning techniques to detect subscription fraud. To build a predictive deep learning model, we used convolutional auto-encoder and deep neural network (DNN) techniques. The fraud detection experimental result showed that the performance evaluation of the experimental result is 99.14% in terms of accuracy measurement using DNN technique on 1048576 voice datasets.