Evaluating the Efficacy of Quantum Machine Learning at Detecting Advance-Fee Fraud Attacks
Farheen Shabbir Shaikh, Tapadhir Das
Advance-Fee Fraud (AFF) is a time-critical financial scam that exploits behavioral urgency, often causing victims to initiate payments within minutes of deceptive communication. Traditional fraud detection systems, which rely on historical transaction patterns, are frequently unable to intervene before the first payment occurs. To address this, this paper investigates the feasibility of Quantum Machine Learning (QML) for rapid AFF detection under Noisy Intermediate-Scale Quantum (NISQ) constraints. For this, we designed a QML framework for AFF detection using behavior-driven temporal features. For evaluation, we compared QNN performance against classical machine-learning baselines using performance metrics and inference latency. Results indicated that even though the QNN achieved efficient performance at detecting AFF attacks with Accuracy, Precision, Recall, and F-1 scores of 0.820, 0.858, 0.793, and 0.824, respectively, it is still currently underperforming in comparison to classical ML models. Additionally, the results also indicate that QML provides slower inference latency compared to traditional ML models, with 2.92 e-03 seconds. These observed results indicate that current QML approaches are best positioned as complementary tools within hybrid detection pipelines, in conjunction with classical ML models.
IEEE SoutheastCon•Feb 20, 2026•0 citations