The ROC curve is a plot of sensitivity versus one minus specificity. Because sensitivity and specificity can be inversely varied by altering the threshold at which a case is categorized as one class or the other, the area under the ROC curve more effectively describes a classification algorithm's discriminatory ability. Stated differently, for a particular implementation of a classification algorithm, the trade-off between sensitivity and specificity is fixed as specified by the ROC curve. To find a different trade-off, one must obtain a different classification algorithm implementation that produces a different ROC curve. This is elaborated upon in figure 4.18.
Was this article helpful?