Application Spotlight: Dirac-3 and CVQBoost push the boundaries of accuracy in detecting financial fraud

All of us have run afoul of credit card fraud at some point in our lives, and financial institutions report billions of dollars lost each year to financial fraud. Traditionally, fraud is detected using rule-based screenings, where people will create a set of conditions that trigger a response if they’re tripped. This can include things like looking at merchant blacklists and travel and seems physically impossible. But these rules are static and easily guessed at by fraudsters, which helps them get around them.

As advanced computing has taken the world by storm, fraud detection has begun to shift to using machine learning algorithms to detect possible fraud. These methods have the advantage of being able to dynamically adapt to changing conditions in the fraud landscape and have already been shown to outperform static, human-created rules for fraud detection, but they still face their own challenges. They struggle in situations where fraud is highly unlikely, but still present, causing the machine to be less sensitive to the fraud that does occur. Additionally, the landscape of variations and patterns that these solvers must navigate can rapidly become extremely complicated, making solutions difficult even for the most powerful classical solvers. Finally, they struggle in situations where deviations are subtle.

QCi collaborator and customer, Dr. Paul Griffin, and his team at Singapore Management University have applied quantum machine learning to the problem of fraud detection. By using QCi’s Dirac-3 quantum optimization machine and proprietary CVQBoost algorithm, they demonstrate results that outperform the best result from a classical method. Most notably, the quantum algorithm was able to significantly improve upon the detection of true positive fraud events when compared to false positives.

Classical algorithms are already relatively good at detecting when no fraud has occurred, exceeding 97% detection across all samples, but when comparing only the ratio of actual fraud cases to the detected positive flags, that accuracy drops to below 75%. False positives are of particular importance because all positive flags would be investigated by a human operator, which costs time and money compared to automated detection. By increasing the accuracy of positive detections, this would reduce the unnecessary workload on human operators. The results using CVQBoost and Dirac-3 show that the detection of positive events rises to 81% accuracy.

Dirac-3, QCi’s flagship quantum optimization machine, is designed to excel at solving optimization problems, and its power can be applied to more than just fraud detection. Its ability to handle high-dimensional problems help it excel at mapping more natural and complex problems. In another study, yet to be published, Dr. Griffin and his team have investigated its uses in optimizing financial portfolio returns.

Interested in learning more about the financial applications of QCi technology?  Join us at Q2B in Tokyo, Japan, for a joint presentation from QCi and Singapore Management University on “CVQBoosting for Improving Credit Card Transaction Fraud Detection” on June 4th.

Read More: 

Loke, B. H. T., Sahoo, N., Guan, B., Xu, M., Verma, D. and Griffin, P. R. (2026). “Improving Credit Card Transaction Fraud Detection Using CVQBoosting.” Proceedings of the 18th International Conference on Agents and Artificial Intelligence - Volume 1: QAIO <https://www.scitepress.org/Link.aspx?doi=10.5220/0014628400004052>