Machine Learning and Econometrics in Credit Card Fraud Detection: An Empirical Analysis

The exponential expansion in the volume of online transactions in the last years has evidenced how the underlying risk of credit card fraud is also rising. Addressing this topic, we can find a potential solution within the use of new computational technologies, as machine learning algorithms, combined with econometric analysis tools. The purpose of this is work is building a Machine Learning model in order to be able to detect credit card fraud with the highest degree of accuracy possible, starting from the analysis of a dataset created with PaySim, a software that simulates credit card transactions based on a real dataset, due to privacy reasons. The empirical analysis was performed through diverse machine learning algorithms, resulting in the selection of a specific type of Random Forest algorithm, the Balanced Random Forest (BRT) algorithm. The selection of this specific algorithm was due to its elevated accuracy in both model building and in handling the severe class imbalances issue emerged in the analysis of the dataset. Resulting from the analysis and model building it is safe to say that utilizing new technologies as machine learning algorithms can bring a great advantage in detecting credit card fraud cases and substantially improve protection from this kind of threat that is everyday more relevant due to the progressive digitalization of our economic system.