Is it possible to make a custom caret package metric which optimize precision at a given recall for binary classification?
A little background: The modelling problem
