How Deutsche Bahn stays ahead of fraudsters using ML on
summary
Deutsche Bahn processes over 100 million online tickets a year (about 200 per minute, 1,000 at peak) and builds an ML fraud stack on SageMaker with API Gateway, Lambda, DynamoDB and a model registry. A prototype runs in weeks: an experimentation environment in one week and a first model in two. Production uses an MLOps pipeline and real‑time DynamoDB features to score up to 1 million transactions/day, with an Explainability Dashboard and Threshold Slider to tune false positives vs fraud loss.
AI-written summary · about
chapters (9)
- 0:00Introduction to fraud prevention
- 6:40Why machine learning for fraud
- 13:15Deutsche Bahn’s fraud challenge
- 18:15Building trust in ML models
- 24:05Evaluating model performance
- 28:04Scaling ML in production
- 33:06Key takeaways and next steps
- 37:45Prototype engagement benefits
- 40:20Getting started with AWS tools
related talks
37mThe Role of Machine Learning in Today’s Payment Ecosystems
16mLeveraging ML to Create a Reliable Anti-Fraud System
39mFraud Detection: Using ML to Identify and Manage Fraudulent Activities
18mDesigning a scalable ML Ops Pipeline: Insights and best practices
13mHarnessing Machine Learning to Combat E-commerce Fraud
36mContinuous Monitoring of Machine Learning Models
27mDetecting fraud faster with Amazon SageMaker, featuring Coinbase
31mCatch more potential online fraud faster with Amazon Fraud Detector