Detecting Digital Ad Fraud by Active Reinforcement Learning

The fraudulent practice of generating false advertising traffic with the aim of unfair profit in the digital advertising market has become increasingly serious. However, discerning whether customer activities are genuine or forged, given the nature of digital information, is a highly challenging task. Current methods for detecting digital advertising fraud rely on expert judgment, machine learning, or logic-based systems, which have limitations in terms of accuracy and adaptability. This paper describes a novel approach that utilizes a reinforcement learning algorithm based on episodic memory and active learning techniques to efficiently detect digital advertising fraud.