Evaluating Classical and Quantum Machine Learning for Credit Card Fraud Detection: Performance and Economic Impact

Outlet Title

2025 Cyber Awareness and Research Symposium (CARS)

Document Type

Conference Proceeding

Publication Date

1-16-2026

Abstract

Credit card fraud detection is an increasingly urgent problem in fintech cybersecurity that requires accurate and scalable solutions against evolving attack patterns. Classical machine learning methods, such as Support Vector Machines (SVM), have shown success; however, challenges with high dimensionality and scalability remain. Quantum computing offers new avenues through quantum parallelism and kernelized methods. In this work, we empirically compare classical and simulated quantum machine learning approaches—SVM, Quantum SVM (QSVM), Variational Quantum Circuits (VQC), and hybrid quantumclassical models—for fraud detection on a balanced synthetic variant of the Kaggle Credit Card Fraud Detection dataset (50% fraudulent, 50% legitimate). Classical SVM achieved F1score of 0.9739 with 0.1 -second training time, while simulated quantum models (Hybrid QSVM and QSVC) achieved F1-scores approaching 1.000 but required over 60 seconds training time—a 600× computational overhead. Economic analysis estimates potential annual savings of 9.7−10M for improved detection accuracy. All quantum experiments were conducted in simulation using Qiskit Aer and PennyLane. While simulated quantum models achieved high accuracy under controlled conditions, these results are optimistic due to random data splits and noiseless simulation environments. Practical deployment will require timeordered validation, evaluation on imbalanced datasets, and future implementation on real quantum hardware with noise considerations.

Link to Full Text

Share

COinS