Real-Time Fraud Detection Using Machine Learning

Financial crime is increasingly facilitated by technology and globalization, demanding advanced IT tools for detection. This paper presents an approach to automate the detection of suspicious activities in real time using Machine Learning (ML) techniques. The approach is designed and evaluated to operate under pragmatic operational constraints inherent to financial institutions, such as extreme class imbalance, stemming from the rarity of fraudulent events relative to legitimate transactions; evolving fraud patterns (concept drift), driven by adversarial adaptation; and significant delays in obtaining verified feedback. The paper aims to elucidate financial transaction monitoring in a data stream context and presents a developed streaming ML pipeline and an experimental testbed.