Artificial Intelligence in Action: Realtime Payment Fraud Detection and Resolution Systems in High Volume Financial Networks

Abstract

This chapter provides an overview of recent research on systems employing artificial intelligence in financial networks, where security, consistency, and timeliness are crucial. Such systems continuously scan high-volume payment operations, targeting the detection of any fraud cases, including both particular attacks and new previously unseen patterns. In contrast to many generic machine learning projects that remain in labs and experimental mode, the successful application of artificial intelligence in already fast-working financial payment systems for large corporations is discussed. We summarize the crucial requirements for the data and the models, as well as particular methodologies and algorithms employed in these holistic systems—starting with behavior, context, time, sequence, and relations modeling, verification of diversified machine learning classifiers' combinations for similar environments, and extending to artificial intelligence explanations, interactive interfaces, and operational process integration. Finally, the software structure of the real-time detection workflow of multiple combined unsupervised models, alongside brief empirical validation, is described.

In most scenarios, from both business and regulatory compliance perspectives, the time component should be taken into account to address operational risks in super real-time—before the current session closes. Before going deeper into discussing the principles and details of fraud detection and response solutions under consideration, key facts need to be highlighted. Operating in a high data traffic domain, financial detection systems must work automatically all the time without human intervention, from detection trigger to operative case conclusion, to protect the area against sophisticated attacks. Moreover, from a corporate customer’s hard-earned reputation point of view, the speed of fraud case resolution is very important.