Q-GALNet With MemAx: A Quantum-Inspired Deep Framework for Scalable Billing Anomaly and Financial Fraud Detection

Abstract

Billing anomaly detectors need to be both credible and scalable such that they facilitate revenue loss avoidance in the banking sector. In this paper, a multi-level fraud detection model is developed in the form of quantum gated adaptive learning network (Q-GALNet) with memory-Addax (MemAx) feature optimisation algorithm. Proposed Q-GALNet applies a quantum-inspired attention mechanism for better feature representation and noise rejection in transactional complex data. The proposed MemAx module offers a dynamic memory-based running selection process to allow for the retention of useful characteristics and some redundant or useless features, thus improving computational efficiency and model generalisation. We instantiate the proposed framework with three benchmark financial datasets, namely PaySim, credit card fraud and IEEE-CIS. Experiments demonstrate that it behaves reliably and performs better than state of the art. The results shows AUC-ROC, AUC-PR, F1-score and recall of 99.60, 98.98, 98.82 and 98.85 respectively. The framework dramatically reduces training time, testing latency, memory usages and total computation cost by 28.8, 37.5, 33.3 and 37.5 respectively These results suggest that the potential hybrid architecture discussed in this work provides a practical and scalable model for real-time detection of billing anomalies and financial fraud across both existing and emerging digital transaction systems.