Telecommunication Fraud Prediction Using Backpropagation Neural Network

Telecommunication fraud prediction in this paper is an interpolation problem that uses the daily telecommunication network services information to analyze and predict the alarm information generated from a detection engine by minimizing false cases. This paper proposed the use of backpropagation neural network (BPNN) to perform telecommunication interpolation based on local telecommunication network services. The backpropagation algorithm is adjusted accordingly to control the speed of obtaining solution. The validation of this method is carried out following a series of experiments to establish the suitable BPNN parameters value for this interpolation problem. 20,000 cases were used and divided suitably into training and testing samples in analyzing the performance of the BPNN network. It is observed that the performance of BPNN in predicting fraud was merely 100% if the threshold value was set to 0.64 and above the predicted value. Consequently, the network model created could be used to analyze fraud risk classification and subsequently, provide contribution to the domain of fraud detection system.