An algorithm developed by University of Italian Switzerland (USI) should in future identify and prevent fraudulent credit card transactions. It was developed as part of a research project financed by the AXA Research Fund. According to information from project supervisor and Head of the Data Science Lab at USI, Prof. Antonietta Mira, the prospects of both AXA and other credit card companies working with the algorithm in future look very promising.
“We have developed a system that works in two phases, the training phase followed by the classification phase”, explains Dr. Bruno Buonaguidi, who worked on the team headed up by Prof. Mira, in a press release. In 2016, he won the corresponding Post-Doctoral Fellowship of the AXA Research Fund.
Initially, the algorithm studies both user behavior of the credit card owner as well as familiar characteristics of reported fraudulent transactions. It calculates a specific threshold value for every individual, which helps to indicate whether a transaction is likely to be fraudulent or can be deemed as safe or regular in nature. A fraud alert is triggered if, for example, the period between two transactions is very tight or if they involve unusually large amounts and are made in different locations than usual.
Information from USI quantifies the annual value of fraud damage for credit card companies at around 25 billion US dollars globally, although this could double by 2025.
According to Antonietta Mira, “the prospects for application by our project partner, as well as other credit card companies, are good”. The algorithm has been tested against real data and was compared with other machine learning methods currently in use across the industry. In so doing, Mira’s team discovered that the percentage of actual fraudulent transactions uncovered is higher than the rate of false positives.