Figure 7

ROC curves. ROC curves of the final classifier, which is constructed using the most frequent and stable results from the cross-validation trials. Parameters for the classifier have been learned from the complete training data set; the resulting classifier has been used to reclassify the training data (solid line) and a set of withheld data. The results from the training data are better than the average cross-validated results since the bin boundaries and probability parameters used in the final classifier come from the entire training data set, not a cross-validation subset.