Classifier performance as a function of the feature set size. The classifier was evaluated in an iterative process where one feature was added at a time. Features were selected according to the ranked feature list (see Table 3), beginning with the best feature. In black the AUC values (y-axis) for the corresponding feature set size (x-axis) are shown. The red line indicates the AUC value of the feature set that was achieved by the feature subset selection approach.