Analytics & AI
Amazon Fraud Detector
- Timebilled
- Bytesnot billed
- Callsbilled
In one line
Predictions are cheap per call; the deployed model holds the meter open.
What gets missed
Billed training compute hours can exceed elapsed training time, because Fraud Detector picks bigger instances than the baseline spec. A deployed model then bills hourly for hosting until you undeploy it.
Find it on your bill
Filter Cost Explorer or your Cost and Usage Report by these usage types. A usage type with no region prefix means us-east-1.
TrainingHrsHostingHrsFraudPrediction-OnlineFraudInsights
Bills the same way
These services turn exactly the same meters, so what you learn here transfers.