Toward Practical Quantum Machine Learning: A Novel Hybrid Quantum LSTM for Fraud Detection

TL;DR

The paper proposes a practical hybrid quantum-classical neural network for credit card fraud detection by integrating a classical LSTM with a variational quantum circuit (VQC). It demonstrates that AngleEmbedding and Strongly Entangling Layers in the quantum module can enrich feature representations, achieving competitive test accuracy while reporting notably fast per-epoch training times (45–65 seconds) on CPU-based simulation. End-to-end training combines classical gradients with quantum gradients via the parameter-shift rule, enabling joint optimization and fair benchmarking against a classical LSTM baseline. The results suggest hybrid quantum-classical approaches can offer meaningful gains in recall and F1 for fraud detection, with practical efficiency and clear directions for scaling to larger datasets and hardware-accelerated implementations.

Abstract

We present a novel hybrid quantum-classical neural network architecture for fraud detection that integrates a classical Long Short-Term Memory (LSTM) network with a variational quantum circuit. By leveraging quantum phenomena such as superposition and entanglement, our model enhances the feature representation of sequential transaction data, capturing complex non-linear patterns that are challenging for purely classical models. A comprehensive data preprocessing pipeline is employed to clean, encode, balance, and normalize a credit card fraud dataset, ensuring a fair comparison with baseline models. Notably, our hybrid approach achieves per-epoch training times in the range of 45-65 seconds, which is significantly faster than similar architectures reported in the literature, where training typically requires several minutes per epoch. Both classical and quantum gradients are jointly optimized via a unified backpropagation procedure employing the parameter-shift rule for the quantum parameters. Experimental evaluations demonstrate competitive improvements in accuracy, precision, recall, and F1 score relative to a conventional LSTM baseline. These results underscore the promise of hybrid quantum-classical techniques in advancing the efficiency and performance of fraud detection systems. Keywords: Hybrid Quantum-Classical Neural Networks, Quantum Computing, Fraud Detection, Hybrid Quantum LSTM, Variational Quantum Circuit, Parameter-Shift Rule, Financial Risk Analysis

Figures (8)

• Figure 1: Model Architecture
• Figure 2: Decomposed quantum circuit used in our QLSTM-based fraud detection model. Each qubit is initialized in the state $\ket{0}$ and undergoes a data-encoding $R_y$ rotation with parameter $x_i$, followed by a trainable $R_z(\theta_i)$. The circuit includes barriers to demarcate the encoding, entanglement, and final transformation stages for clarity. The entanglement layer is implemented using staggered CNOT gates between adjacent qubits to ensure local entanglement. This is followed by a second set of trainable rotations $R_y(\phi_i)$ and $R_z(\psi_i)$. All qubits are measured in the computational (Z) basis. This diagram is a simplified decomposition of the PennyLane StronglyEntanglingLayers template used in the model.
• Figure 3: Flow diagram summarizing the training process: data loading, forward pass through the hybrid model, gradient evaluation (integrating classical gradients and quantum gradients via the parameter-shift rule as detailed in Methodology Section), and optimizer update followed by metric logging.
• Figure 4: Combined training performance over 80 epochs. The left subfigure shows accuracy, while the right subfigure shows loss.
• Figure 5: Time Per Epoch for Training. Most epochs require 45--65 seconds, showcasing the model's relatively fast convergence despite incorporating quantum layers.