You are creating a classification model to identify credit card fraud for a banking company using automated machine learning in Azure Machine Learning. The training dataset is highly imbalanced.

Which primary metric should you use to evaluate the model?