ConvNets for Fraud Detection analysis

Fraud activity is a big concern for telecom companies. The advances in technology and system information have significantly increased fraud activities, which can have negative impacts on revenue gains and services quality. Therefore, there is an urgent need for telecom companies to develop efficient algorithms that detect early potential frauds and/or prevent them. In this paper, we used deep learning techniques as an effective method to detect fraudsters in mobile communications. Fraud datasets from the customer details records (CDR) of a real mobile communication carrier were used and learning features were extracted and classified to fraudulent and non-fraudulent events activity. Different experiments were performed to evaluate the performance of our proposed model. We found that deep convolution neural networks (DCNN) technique outperformed other traditional machine learning algorithms (Support vector machines, Random Forest and Gradient Boosting Classifier) in term of accuracy (82%) and training duration. Thus, the use of this model can reduce the cost related to illegal use of services without payment.