Speaker
Description
Financial fraud detection is a highly imbalanced and dynamic learning problem that challenges traditional machine learning models. Quantum machine learning (QML) offers new representational paradigms capable of exploring high-dimensional feature spaces in ways substantially different from classical systems, leveraging quantum superposition and entanglement. In this study, we present a comprehensive evaluation of quantum and hybrid QML architectures, specifically Quantum Autoencoders, Hybrid Diffusion Time Estimation, and Hybrid Variational Autoencoders, applied to real-world banking data encompassing millions of transactions. By combining deterministic preprocessing, feature selection, and balanced validation protocols, we assess the comparative advantages of quantum-enhanced inference under realistic fraud detection conditions. We implement quantum models via classical simulations. Results indicate that even if current NISQ hardware constraints limit scalability, quantum–classical hybrids can yield measurable improvements in anomaly sensitivity and calibration over standard baselines for specific data regimes. Classical models generally remain more consistent and robust overall. This study provides one of the first end-to-end empirical analyses of QML in operational financial contexts, illustrating both its feasibility and open research challenges.
| Sessions | Quantum Machine Learning: |
|---|---|
| Invited | No |