Premium Rate Services Fraud Detection

Premium Rate Services (PRS) destination fraud occurs when telephone calls are made to high-cost (premium) destinations through fraudulent means. Early detection and termination of such calls can significantly reduce the cost they incur. We investigate various features of such calls to determine which can be used to quickly predict and terminate them. This is achieved through feature selection and clustering analyses followed by fraud detection using a Classification and Regression Tree (CART) model. Additionally, we propose the use of costs/benefits to evaluate the model’s performance through the consideration of an adjusted decision tree cost function at the training level. This approach is compared to the traditional decision tree to highlight its advantages.