ROCDiagramm

ROCDiagramm, usually written as ROC-Diagramm or ROC-Kurve in German, is a graphical tool for evaluating the performance of binary classifiers. It shows the relationship between the true positive rate (TPR, also called sensitivity) and the false positive rate (FPR, which equals 1 minus specificity) across a range of decision thresholds. The diagram is widely used in statistics, medicine, radiology, and machine learning to assess how well a model can distinguish between two classes. The area under the curve (AUC) summarizes the overall discriminatory ability of the classifier, with 0.5 indicating no discrimination and 1.0 indicating perfect discrimination.

Construction: A classifier outputs a score or probability for each instance. By varying the threshold, TPR and FPR are computed for each threshold and plotted to form the ROC curve. The AUC is commonly estimated using the trapezoidal rule. Comparisons between ROC curves can be tested statistically, for example with methods like DeLong’s test.

Interpretation: A curve nearer to the top-left corner indicates better performance, as it achieves high TPR with low FPR. The diagonal line from (0,0) to (1,1) represents random guessing. ROC diagrams are threshold-independent, making them useful for comparing models without selecting a specific threshold.

Applications and limitations: ROC diagrams are standard in medical diagnostics and in machine learning model evaluation to compare performance and select thresholds. Limitations include that AUC does not reflect probability calibration and can be misleading with highly imbalanced data; in such cases, precision–recall analyses may be more informative.