Overlays the ROC curves of three fraud classifiers so their areas under the curve can be compared at a glance; the gradient-boosting model bows hardest toward the top-left and dominates the others. A chance-level reference line marks the no-skill baseline.