A Quantum Leap In Fraud Detection Accuracy Using Graph-based Features

Big Data LDN 2022, 21 Sept, 2:20pm, Modern Analytics Theatre
Speaker:
Richard Henderson, TigerGraph
Graph databases have been gaining ground as an alternative to relational databases, with significant potential value, but a simple route to that value hasn't always been obvious. This talk describes the latest results of using a high-performance graph database plus two data-independent algorithms to provide significant (20%-50%) uplift to existing ML models, with examples of how it went into production at some of the biggest banks in the world, in a matter of weeks, with no significant changes needed to data-feeds, pipelines, or ML model-types except to add the graph-based features. It will cover how anyone can test this in their own ML pipeline with minimal effort or new expertise needed, using a freely-available packaged toolkit.