Data Mining Solutions for Fraud Detection in Credit Card Payments

We describe an experimental approach to design a Fraud Detection system using supervised Machine Learning (ML) methods such as decision trees and random forest. We believe that such an approach allows financial institutions to investigate fraudulent cases efficiently in terms of accuracy and time. When ML methods are applied to imbalance problems such as Fraud Detection, the outcomes of decision models must be accurately calibrated in terms of predicted fraud probabilities. The use of different ML models allows practitioners to minimize the risks and cost of the solutions. We discuss the main results obtained in our experiments on the benchmark problems.