In the context of developing a machine learning model to classify potentially fraudulent transactions, where the goal is to rank models based on both precision and recall, which evaluation metric would you suggest?