Describe how you might use a ROC curve to justify how you pick an value. Assume only a single test will ultimately be conducted on the actual data, so you don't have to correct for multiple tests.