Fraud & Risk
A Fraud Detection System Based on Customized ML Models
This presentation begins with a high-level overview of Microsoft's Fraud Detection System, followed by discussion of how pre- and post-purchase analysis is modeled and how customer behavior can be summarized using vectors. Fraud islands and concept drift are then defined and described, with sample equations included to help illustrate. The review concludes with a comparison of long-term and short-term sequential models, and their impact on model architecture.