Comparative Performance Analysis of Quantum Machine Learning Architectures for Credit Card Fraud Detection

TL;DR

Credit card fraud detection is increasingly challenging, motivating an empirical comparison of three QML architectures—VQC, SQNN, and EQNN—across two real-world datasets with multiple feature maps and ansatz configurations. The authors use ANOVA to validate performance differences and assess robustness under five quantum noise channels, finding VQC and SQNN generally outperform EQNN, with VQC achieving the highest F1-score on the European dataset (up to 0.88) for certain feature-map/ansatz combinations. Entanglement-enabled feature maps (ZZ and Pauli) improve learning when paired with expressive ansatz, while simpler maps (Z) can be competitive for less expressive models or noisier settings. The work demonstrates practical QML viability on near-term devices, guiding architecture and encoding choices for robust quantum fraud detection and informing future research on deeper or noise-aware designs.

Abstract

As financial fraud becomes increasingly complex, effective detection methods are essential. Quantum Machine Learning (QML) introduces certain capabilities that may enhance both accuracy and efficiency in this area. This study examines how different quantum feature maps and ansatz configurations affect the performance of three QML-based classifiers, the Variational Quantum Classifier (VQC), the Sampler Quantum Neural Network (SQNN), and the Estimator Quantum Neural Network (EQNN), when applied to two non-normalized financial fraud datasets. Different quantum feature map and ansatz configurations are evaluated, revealing distinct performance patterns. The VQC consistently demonstrates strong classification results, achieving an F1-score of 0.88, while the SQNN also delivers promising outcomes. In contrast, the EQNN struggles to produce robust results, emphasizing the challenges presented by non-standardized data. Statistical validation using ANOVA confirms the significance of observed performance differences. Additionally, robustness tests on the best-performing models under five quantum noise types show that they maintain competitive performance, supporting their practical applicability. These findings highlight the importance of careful model configuration in QML-based financial fraud detection. By showing how specific feature maps and ansatz choices influence predictive success, this work guides researchers and practitioners in refining QML approaches for complex financial applications.

Figures (20)

• Figure 1: Overview of the proposed methodology. Two benchmark datasets undergo initial preprocessing, after which three QML models are evaluated. Each model is tested with three distinct feature map architectures, and each selected feature map is further assessed using four different ansatz configurations.
• Figure 2: Fundamental architecture of a QML model. It involves transforming classical data into quantum states $|\psi_i\rangle$, processing them through an ansatz $\mathbf{U}(\theta)$, and optimizing its parameters using a classical optimizer based on a loss function until convergence is achieved iteratively. The final quantum state is measured to generate predictions.
• Figure 3: Graphical representation and mathematical expressions of selected single and two-qubit quantum gates.
• Figure 4: Architectures of the four ansatz variants utilized in this study, demonstrated on seven qubits: a) Real Amplitudes, b) Efficient SU2, c) Two Local, and d) Pauli Two Design, highlighting their structural differences and configurations.
• Figure 5: Architecture of the VQC model, it consists of a feature map, the $U$ operator is then applied in a measurement to obtain the classical output $\hat{y}$. A classical optimizer iteratively updatethe parameters $\theta$ to minimize the loss function $J(y; \hat{y})$.