Amazon Fraud Detector
Managed fraud-risk scoring built on Amazon's experience.
❓ What is it?
A managed service that trains fraud-detection models on your historical event data (plus fraud patterns learned at Amazon) and returns real-time risk scores with rule-based decisions for events like signups and payments.
💡 Why does it exist?
Fraudsters adapt faster than hand-written rules. A learned risk score plus adjustable business rules catches evolving patterns while keeping the block/allow policy in your hands.
⏱️ When should you use it?
Use it for online transaction fraud, fake account creation, promotion abuse, and card-testing detection — cases with labeled historical events and a need for millisecond decisions in the request path.
🗺️ Where does it fit?
Synchronously inside checkout/signup flows: the backend calls GetEventPrediction, receives a score and rule outcome, and branches (approve, review, deny) before completing the action.
🔌 How do you integrate it?
Upload labeled events to S3, train a model type suited to the use case (e.g. online fraud insights), combine it with decision rules in a detector, and call the prediction API per event.
🧩 Commonly integrated with
🎯 Exam angle (AIF-C01)
- "Detect fraudulent transactions/accounts without building models" → Fraud Detector; generic anomaly detection in metrics belongs to CloudWatch/Lookout instead.
- Scores feed RULES you control — the service does not hard-code block decisions for you.