Fraud is evolving. Phishing schemes, payment fraud, identity spoofing, large scale automated attacks,... the methods fraudsters use are constantly shifting, and they adapt quickly whenever new countermeasures are put in place. Traditional detection systems, built on fixed rules and static models, are struggling to keep up.
The FraudShift chair was created to address this challenge. Our research focuses on fraud analytics: a field where statistics and machine learning come together to detect suspicious patterns in complex, real world data. The goal is not just to build more accurate detection systems, but systems that are reliable, cost-efficient, and robust enough to remain effective as the fraudsters adapt the techniques.
Fundamental Research
A core part of our mission is to advance the public body of knowledge on fraud detection. All findings are published as open-access scientific articles, ensuring that our work contributes to the broader academic and professional community, not just to a single organization or product.
Understanding How Fraudsters Adapt
One of the key challenges in fraud detection is that fraudsters do not stand still. When a countermeasure is deployed, they observe it and adjust their behaviour accordingly. FraudShift will studie these adaptation mechanisms: modelling how fraudulent strategies evolve over time in response to detection systems. Understanding this dynamic is essential to staying one step ahead rather than always catching up.
Detecting Weak Signals Early
Fraud rarely announces itself with a clear alarm. More often, it starts as a subtle anomaly, a small deviation in behavior that, left unnoticed, grows into a serious incident. We develop methods to identify these weak signals in data environments that are continuously changing, so that problems can be flagged before they escalate.
Adaptive, Cost-Aware Systems
In practice, fraud detection is never a clean binary problem. Data is uncertain, incomplete, or contaminated. And the costs of errors are not symmetrical, missing a fraud case and incorrectly flagging a legitimate transaction carry very different consequences. FraudShift will design systems that handle this complexity: adjusting continuously to new patterns, managing uncertainty intelligently, and taking the real-world costs of decisions into account.
The Research Team
FraudShift is led by Prof. Tim Verdonck and Prof. Jakob Raymaekers, both from the Department of Mathematics at the University of Antwerp. Prof. Verdonck is a specialist in statistical machine learning and anomaly detection in financial and insurance data, with over 110 scientific publications to his name. Prof. Raymaekers brings expertise in robust statistical methods for high-dimensional data, and is a co-developer of the open source package RobPy. Together, they supervise a team of doctoral researchers from the University of Antwerp.