Quantum-inspired reinforcement learning for adaptive fraud detection in large-scale mobile ad networks

This paper introduces Quantum-Inspired Reinforcement Learning (QIRL) as a novel approach to detecting fraudulent activities in large-scale mobile advertising networks, addressing critical limitations of current detection methods. The proposed architecture integrates quantum-inspired computing with reinforcement learning, leveraging quantum-inspired state representations that encode advertisement interaction data into amplitude-phase pairs. The method was evaluated on a dataset of 7.2 million advertisement events with a highly imbalanced class ratio of 16.26:1. QIRL achieves an accuracy rate of 89.7%, precision of 76.9%, and recall of 74.8%, substantially outperforming traditional methods. The approach reduces false positive rates by 23.5% compared to baseline algorithms and demonstrates outstanding adaptability to evolving fraud strategies, including click injection, click spamming, install attribution fraud, and SDK spoofing. Quantum-inspired reinforcement learning represents a significant advancement in mobile advertising fraud detection, offering superior detection capabilities and dynamic adaptation to emerging fraud patterns. This approach can help recover significant advertising revenue and restore confidence in mobile advertising infrastructure by providing more accurate and adaptive fraud detection capabilities that minimize false positives while maintaining high detection rates.