Artificial intelligence and machine learning have been truly reshaping the payment industry, especially regarding fraud prevention. Recently, many successful machine learning applications have been developed for data-mining programs that learn to detect fraudulent credit card transactions.
In order to be in step with the competition and follow the most current trends, Mercury Processing Services International has too recognized the need to introduce improved machine learning algorithms into the fraud detection process. To realize this, Mercury Processing Services International started the collaboration with the Faculty of Electrical Engineering and Computing from Zagreb, with the long-term plan to improve the scoring model with machine learning techniques.
After using statistical model based on a simple machine learning algorithm, Mercury Processing Services International is now working on a more sophisticated machine learning model that will be regularly retrained and maintained internally.
The first part that is finished consisted of determining the machine learning algorithms applicable for card fraud detection and prevention, apply algorithms on defined data set and analyse the efficiency of fraud detection, compare the reselected results of the current Rule engine (that uses current scoring module and business rules) and new algorithms.
Currently, Mercury Processing Services International is in the process of developing the random forest model over a larger set of data on the test system.
Nataša Benčić, Senior Product Expert from Mercury Processing Services International, is one of the people working on the project. She said that some of the main challenges the development process is facing are concerning feature engineering, unbalanced class sizes, scalability, and fraud patterns changes.
Nevertheless, Mercury Processing Services International is looking forward to overcome the challenges and continue with the other phases towards the final goal.