Case Study: PayPal achieves 6% higher fraud-detection accuracy and 6X faster model development with H2O.ai Driverless AI

Driving Away Fraudsters at Paypal

PayPal, a global online payments leader, faced persistent and evolving fraud from colluding buyers and sellers that threatened its buyer and seller protection programs. To stay ahead of sophisticated fraud networks and speed up model development, PayPal partnered with H2O.ai and used H2O Driverless AI to augment its existing fraud-detection efforts.

PayPal combined graph analytics (neo4j + node2vec) and expert features, then applied H2O Driverless AI for automated feature engineering and model training; the top five features generated by H2O Driverless AI outperformed ten years of expert features. The result: model accuracy rose from 0.89 to 0.947 (nearly a 6% gain), training ran about 6× faster on IBM Power GPU hardware, and PayPal continues to expand use of H2O.ai to detect collusion fraud more effectively.

PayPal

Venkatesh Ramanathan

Senior Data Scientist