The pervasive threat of online fraud significantly impacts businesses, necessitating a robust end-to-end strategy for the detection and prevention of new account fraud, account takeovers, and suspicious payment transactions. Crucial to the success of a fraud detection and prevention system is the capability to identify fraud in close proximity to its occurrence. It is imperative that the system not only detects fraud effectively but also promptly alerts end-users, empowering them to take immediate action and mitigate further abuse.

To tackle these challenges, the adoption of real-time machine learning becomes essential. However, this introduces a set of hurdles, including ensuring the freshness of features in real-time, maintaining feature consistency during training and inference, backfilling historical data for feature regeneration, ensuring point-in-time correctness, and managing the complexity of streaming systems.

In this session, we will illustrate a solution that leverages apache kafka and streaming database to construct a real-time feature pipeline effectively addressing these challenges. Key highlights include:

Building a low-latency streaming feature pipeline utilizing streaming SQL.
Providing a consistent feature pipeline for both training and serving purposes.
Establishing a simplified, single-box system that is easy to manage and operate.

Join us to explore a practical demonstration of how this real-time feature pipeline can revolutionize fraud detection and prevention in the ever-evolving landscape of online security.