The Precision-Recall Plot Is More Informative than the ROC Plot When Evaluating Binary Classifiers on Imbalanced Datasets

A representation and interpretation of the area under a receiver operating characteristic (ROC) curve obtained by the "rating" method, or by mathematical predictions based on patient characteristics, is presented. It is shown that in such a setting the area represents the probability that a randomly chosen diseased subject is (correctly) rated or ranked with greater suspicion than a randomly chosen non-diseased subject. Moreover, this probability of a correct ranking is the same quantity that is estimated by the already well-studied nonparametric Wilcoxon statistic. These two relationships are exploited to (a) provide rapid closed-form expressions for the approximate magnitude of the sampling variability, i.e., standard error that one uses to accompany the area under a smoothed ROC curve, (b) guide in determining the size of the sample required to provide a sufficiently reliable estimate of this area, and (c) determine how large sample sizes should be to ensure that one can statistically detect differences in the accuracy of diagnostic techniques.

Diagnostic systems of several kinds are used to distinguish between two classes of events, essentially "signals" and "noise". For them, analysis in terms of the "relative operating characteristic" of signal detection theory provides a precise and valid measure of diagnostic accuracy. It is the only measure available that is uninfluenced by decision biases and prior probabilities, and it places the performances of diverse systems on a common, easily interpreted scale. Representative values of this measure are reported here for systems in medical imaging, materials testing, weather forecasting, information retrieval, polygraph lie detection, and aptitude testing. Though the measure itself is sound, the values obtained from tests of diagnostic systems often require qualification because the test data on which they are based are of unsure quality. A common set of problems in testing is faced in all fields. How well these problems are handled, or can be handled in a given field, determines the degree of confidence that can be placed in a measured value of accuracy. Some fields fare much better than others.