A bank wants to detect fraudulent transactions in real time. What type of inference should they use?

Real-time inference

Explanation

Real-time inference processes data instantly making it ideal for fraud detection where immediate action is required.  Batch inference is used for processing large datasets at scheduled intervals.  Asynchronous inference is suitable for large payloads but not for real-time needs.  On-demand inference is for infrequent queries rather than continuous fraud monitoring.