A machine learning team wants to detect bias in a binary classification model before deployment. They use SageMaker Clarify. Which type of bias metric should they compute to understand whether the model treats different demographic groups unfairly in predictions?