A machine learning specialist must produce near-real-time probabilities that a credit card transaction is fraudulent. How should this problem be formulated?

Binary classification is the correct formulation because the problem requires predicting one of two outcomes: fraudulent or not fraudulent. This is a classic binary decision. Streaming classification describes the method of processing data as it arrives, not the problem formulation itself. While the solution might involve streaming data, the core task is still binary classification. Multi-class classification is incorrect because there are only two possible outcomes, not more than two. Regression is incorrect because the problem asks for a probability (a categorical outcome), not a continuous numerical value.