A STEP-BY-STEP GUIDE TO BUILDING MACHINE LEARNING-BASED MODELS
In the age of AI, traditional rules-based fraud detection solutions are no longer sophisticated enough to catch fraudsters (who are constantly at the cusp of the technological curve). In addition, they often create work for teams who must manually review potential fraud since they are not precise enough.
Make the shift to machine learning-based models with this guidebook, which includes:
Kevin has nearly 10 years of experience in financial services and technology, having previously worked at Oracle and Merlon Intelligence, where he focused on how AI and machine learning could help solve complex challenges within financial crime compliance. He is currently part of the financial services focused sales team across the Eastern United States and Canada at Dataiku.
Harizo has a background in mathematics and computer science and holds a PhD in Computational and Applied Mathematics from the University of Lille. He works on the R&D team at Dataiku, focusing on technical ecosystem integrations, particularly the challenges of enterprise-grade deployments (security, availability, and scalability).
Between compliance regulations, fast transactions, and innovative fraudsters, tackling fraud will never be easy, but it doesn't have to be impossible. This guidebook offers a concrete step-by-step path to a machine learning-based fraud detection POC, and also contains:
Evaluating anomaly detection is a particularly fine balance. False negatives can be detrimental, of course, but on the other hand, a system that identifies too many false positives is of almost no use either. In a real-time system, there is no room for a second review of any potential anomalies if the system is producing too much noise. Banking traditionally leverages rules-based fraud detection models, which are insufficient and often result in false-positive rates that exceed 90%.
And in banking use cases, false positives in anomaly detection could destroy the trust users place in an organization (think of the frustration, for example, if your bank was constantly blocking your funds due to false-positive fraud detection). The solution is to spend the time upfront on feature engineering to make sure there aren’t too many false negatives or positives and continuing to iterate, refine, and make improvements even after the model is in production.
The main challenge when it comes to detecting fraud is that data is imbalanced -with so many non-fraudulent transactions, building and training a machine learning model can be challenging. This guidebook walks step-by-step through all you need to know to start using machine learning based fraud detection techniques for ultimately more effective and less error-prone systems.