The correct answer is Azure Stream Analytics.

Azure Stream Analytics is a managed, low latency streaming analytics service that is designed to evaluate event and transaction streams in real time and produce immediate results for scenarios such as fraud detection. It provides a SQL like query language for windowed aggregations and pattern detection and it integrates directly with Event Hubs, IoT Hub, and other ingestion services to process transactions as they arrive.

Azure Stream Analytics also supports calling out to external model endpoints or user defined functions so you can apply trained machine learning models for scoring in the streaming pipeline with minimal operational overhead. That makes it a good fit when you need immediate, continuously running fraud detection rather than periodic batch scoring.

Cloud Dataflow is not the right choice because it is a Google Cloud service built around Apache Beam and it is not the native Azure streaming service for this scenario.

Azure Machine Learning is focused on training, managing, and hosting models and it can serve real time endpoints but it is not itself a managed streaming analytics engine for applying models across a continuous event stream.

Azure Databricks can perform streaming analytics with Structured Streaming and it can apply models, but it requires more engineering and cluster management and it is generally more complex than using a purpose built managed streaming engine when you need low latency, production ready fraud detection.

Full AWS Practitioner Certification Question