The user uploads the data into an Amazon Simple Storage Service (Amazon S3) bucket to be used for training both a custom model in Amazon SageMaker and a model in Amazon Fraud Detector. The custom fraud detection model is trained in SageMaker, model artifacts are saved in an S3 bucket with server-side encryption enabled, and the model is deployed using a SageMaker model endpoint. A fraud detection model is created in Amazon Fraud Detector. The custom SageMaker model is exported to Amazon Fraud Detectorusing the deployed SageMaker endpoint created in step 2. A detector that combines scores from both the SageMaker model and the Amazon Fraud Detector model is created, which has multiple decision rules. At the time of inference, the incoming request coming from the user is received and sent to AWS Lambda for processing and invoking the fraud models. A copy of the data can be also saved on anS3 bucket through Amazon Kinesis streaming services. When AWS Lambda receives the data to use for inference, it invokes the Amazon Fraud Detector endpoint and receives the prediction. The prediction output is communicated back downstream by AWS Lambda to the user or other downstream applications, as re
