Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Mixture-of-Experts Framework for Practical Hybrid-Quantum Models in Credit Card Fraud Detection

Rodrigo Chaves, Kunal Kumar, Bruno Chagas, Rory Linerud, Brannen Sorem, Javier Mancilla, Bryn Bell

TL;DR

Improvements in fraud detection performance and a possible trade-off of fraud and nominal detections with a reduction in false positives at the cost of a small reduction in fraud detections are revealed.

Abstract

This paper investigates whether hybrid quantum-classical machine learning can deliver practical improvements in financial fraud detection performance for card-based and other payment transactions. Building on a Guided Quantum Compressor architecture, the approach integrates an autoencoder, a variational quantum circuit, and a classical neural head, and then embeds this hybrid model into a mixture-of-experts framework including a state-of-the-art gradient-boosted tree classifier. Using a European credit card dataset with severe class imbalance, the routed hybrid architecture achieves average precision scores of $0.793\pm0.085$ compared to $0.770\pm0.065$ of XGBoost on 3 repeated 5-fold cross-validation benchmarks. Precision and recall comparisons reveals a possible trade-off of fraud and nominal detections with a reduction in false positives at the cost of a small reduction in fraud detections. The improvements are achieved while adding only 7 to 21 minutes of extra inference time depending on the choice of hyperparameters. These results indicate that selectively routing transactions to quantum-classical models can enhance fraud detection while remaining compatible with the latency and operational constraints of modern financial institutions.

A Mixture-of-Experts Framework for Practical Hybrid-Quantum Models in Credit Card Fraud Detection

TL;DR

Improvements in fraud detection performance and a possible trade-off of fraud and nominal detections with a reduction in false positives at the cost of a small reduction in fraud detections are revealed.

Abstract

This paper investigates whether hybrid quantum-classical machine learning can deliver practical improvements in financial fraud detection performance for card-based and other payment transactions. Building on a Guided Quantum Compressor architecture, the approach integrates an autoencoder, a variational quantum circuit, and a classical neural head, and then embeds this hybrid model into a mixture-of-experts framework including a state-of-the-art gradient-boosted tree classifier. Using a European credit card dataset with severe class imbalance, the routed hybrid architecture achieves average precision scores of compared to of XGBoost on 3 repeated 5-fold cross-validation benchmarks. Precision and recall comparisons reveals a possible trade-off of fraud and nominal detections with a reduction in false positives at the cost of a small reduction in fraud detections. The improvements are achieved while adding only 7 to 21 minutes of extra inference time depending on the choice of hyperparameters. These results indicate that selectively routing transactions to quantum-classical models can enhance fraud detection while remaining compatible with the latency and operational constraints of modern financial institutions.
Paper Structure (18 sections, 23 equations, 2 figures, 5 tables)

This paper contains 18 sections, 23 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Theoretical background and methodology
  3. Autoencoders
  4. Angle encoding
  5. Variational Quantum Circuit (VQC) Classifier
  6. Feedforward Neural Networks
  7. Temperature calibration
  8. Combined Model
  9. Mixture-of-experts
  10. Experts training
  11. Router training
  12. Experiments
  13. European Credit Card Dataset
  14. Preprocessing
  15. Validation and Testing
  16. ...and 3 more sections

Figures (2)

  • Figure 1: Circuit used in the architecture as the classifier for 4 qubits with $n$ layers.
  • Figure 2: Validation procedure used during experimentation. The dataset is preprocessed using a MinMax scaler to be further split in train, tests, and holdout sets. Train is used to train the model, test is used to train the router, and holdout to evaluate the model's performance.