Major CEX Upgrades Fraud Detection System with Machine Learning and Rule Engine, Cuts Response Time to Hours

Gate News message, April 23 — A major centralized exchange announced an overhaul of its anti-fraud system by integrating machine learning models with rule-based engines, implementing a dual-track strategy where models handle long-term defense and rules enable rapid response. The unified framework creates a feedback loop: rules capture emerging fraud patterns and feed data back to retrain models, continuously strengthening overall defense capabilities.

The exchange restructured its data infrastructure, automated schema evolution, and introduced notebook-based analytical tools to shift rule creation from manual processes to data-driven automation. Rule backtesting performance improved over 10-fold, while overall response time was reduced from days to hours. Machine learning recommendations for parameters also help lower false positive rates, minimizing impact on legitimate users.

The exchange plans to advance event-driven automated rule generation and explore converting efficient rules into model features with a single click, further moving toward fully automated risk management systems.