Receiver operating characteristic curve (ROC) for (4). Training set is represented by white points and random classifier by black points. TPF (true positive fraction)= sensitivity and FPF (false positive fraction) = 1-specificity, for different thresholds of class function (between −1.0 and ).
