Adversarial Learning in Real-World Fraud Detection: Challenges and Perspectives

What is it about?

As machine learning plays a crucial role in many applications and is indispensable for many scientific, economic, and governmental activities, we should not neglect the immense risks of fraudulent activity, and we need to reinforce such systems with secure learning algorithms and practices. The necessity of operating machine learning models in adversarial environments, where an adversary actively works to have the implemented model behave differently from what it was defined for, led to the creation of a new research field called Adversarial Machine Learning (AML). Over the last two decades, adversarial machine learning has become a research topic of increasingly growing interest, mainly due to the significant initial results obtained in the field of image recognition. Despite the successful application of adversarial techniques to image recognition, generalizing them to other applications and domains is neither trivial nor obvious. This can have severe implications, as it leaves us oblivious to the effective threats that could hinder other applications, thus making it impossible to defend against them. In this paper, we consider the problem of applying adversarial machine learning techniques to fraud detection to ensure the robustness of online transaction systems against hostile attacks, describe how attacks against fraud detection systems differ from other applications of adversarial machine learning and propose several interesting directions to bridge this gap.

Featured Image

Photo by Markus Spiske on Unsplash

Photo by Markus Spiske on Unsplash

Why is it important?

Understanding adversarial attacks against fraud detection systems is a crucial step toward a better understanding of the threat and designing the necessary defenses. This is important for two main reasons: 1) Machine learning plays a crucial role in ensuring the safety of online payments. Successful attacks may destroy the users' trust in such systems, leading to a significant blow to our economies as a whole 2) Fraud Detection is known to be an interesting laboratory for machine learning, as it presents multiple challenges common with various other applications. For this reason, understanding the threat of adversarial attacks in fraud detection can lead to a better understanding of many other applications as well.

Perspectives