AUC-Wert
The AUC Score, or Area Under the Receiver Operating Characteristic Curve, is a common metric used to evaluate the performance of binärer Klassifikation Modelle. Die Receiver Operating Characteristic (ROC) curve itself is a graphical representation that illustrates the trade-off between the true positive rate (sensitivity) and the False Positive Rate (1-Spezifität) bei verschiedenen Schwellenwerten.
Der AUC-Wert quantifiziert die Gesamtfähigkeit des Modells, zwischen positiven und negativen Klassen zu unterscheiden. Er reicht von 0 bis 1, wobei ein Wert von 0,5 keine Unterscheidung bedeutet (ähnlich wie zufälliges Raten), während ein Wert von 1 eine perfekte Unterscheidung zwischen den Klassen anzeigt. Ein höherer AUC-Wert deutet auf ein besseres Modell hin.
To compute the AUC Score, the first step is to generate the ROC curve by varying the threshold for classifying instances as positive or negative. For each threshold, the true positive and false positive rates are calculated, resulting in a curve that plots these rates against each other. The area under this curve is then calculated using numerische Integration Methoden.
One of the advantages of the AUC Score is that it remains unaffected by class imbalance, making it a robust measure in scenarios where one class may significantly outnumber the other. However, it is important to note that while the AUC Score provides a good summary of Modellleistung, it does not convey information about the specific thresholds at which a model operates best or how the model performs at individual decision thresholds.