The image shows a pipeline workflow for live fraud prediction using Oracle Compute Cloud@Customer, Kafka, and a Morpheus pipeline in a Docker container in three swinlanes stacked one below the other.

Host Environment: The flow starts downwards from Transaction data (validation.csv) to the Python producer located in the Host environment.

• On the right, "Live fraud prediction" is written next to an icon of a person.

Kafka: The flow continues downwards to Kafka topic: INPUT (gnn_fraud_input) located in the Kafka swimlane.

• In the middle, under "Host Environment" and above "Morpheus Pipeline," this section represents input and output messaging streams.
• On the left, "" is shown, with an "Input stream" label.

Morpheus Pipeline (Docker Container): The workflow continues from INPUT (gnn_fraud_input) to the Kafka source located in the Morpheus pipeline swimlane.

• Kafka source flows to the Deserialize.
• Deserialize connects to Graph construction.
• Graph construction connects to GNN interface (GraphSAGE).
• GNN interface (GraphSAGE)" (neural network icon) connects to Classification (XGBoost).
• Classification (XGBoost) to Serialize.
• Serialize connects to Kafka sink.

Kafka sink connects back Kafka topic: OUTPUT (gnn_fraud_output) in the Kafka swimlane.

Kafka topic: OUTPUT (gnn_fraud_output) connects to Python consumer in the host environment swimlane.

Python consumer connects to Live fraud prediction in the host environment.