How do you evaluate model performance when dealing with highly imbalanced clinical classification datasets?
