Dataset Information

Area under precision-recall curves for weighted and unweighted data.

ABSTRACT: Precision-recall curves are highly informative about the performance of binary classifiers, and the area under these curves is a popular scalar performance measure for comparing different classifiers. However, for many applications class labels are not provided with absolute certainty, but with some degree of confidence, often reflected by weights or soft labels assigned to data points. Computing the area under the precision-recall curve requires interpolating between adjacent supporting points, but previous interpolation schemes are not directly applicable to weighted data. Hence, even in cases where weights were available, they had to be neglected for assessing classifiers using precision-recall curves. Here, we propose an interpolation for precision-recall curves that can also be used for weighted data, and we derive conditions for classification scores yielding the maximum and minimum area under the precision-recall curve. We investigate accordances and differences of the proposed interpolation and previous ones, and we demonstrate that taking into account existing weights of test data is important for the comparison of classifiers.

SUBMITTER: Keilwagen J

PROVIDER: S-EPMC3961324 | biostudies-literature | 2014

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Area under precision-recall curves for weighted and unweighted data.

Keilwagen Jens J Grosse Ivo I Grau Jan J

PloS one 20140320 3

Precision-recall curves are highly informative about the performance of binary classifiers, and the area under these curves is a popular scalar performance measure for comparing different classifiers. However, for many applications class labels are not provided with absolute certainty, but with some degree of confidence, often reflected by weights or soft labels assigned to data points. Computing the area under the precision-recall curve requires interpolating between adjacent supporting points, ...[more]

PMID: 24651729

Similar Datasets

Project description:BackgroundIncremental value (IncV) evaluates the performance change between an existing risk model and a new model. Different IncV metrics do not always agree with each other. For example, compared with a prescribed-dose model, an ovarian-dose model for predicting acute ovarian failure has a slightly lower area under the receiver operating characteristic curve (AUC) but increases the area under the precision-recall curve (AP) by 48%. This phenomenon of disagreement is not uncommon, and can create confusion when assessing whether the added information improves the model prediction accuracy.MethodsIn this article, we examine the analytical connections and differences between the AUC IncV (ΔAUC) and AP IncV (ΔAP). We also compare the true values of these two IncV metrics in a numerical study. Additionally, as both are semi-proper scoring rules, we compare them with a strictly proper scoring rule: the IncV of the scaled Brier score (ΔsBrS) in the numerical study.ResultsWe demonstrate that ΔAUC and ΔAP are both weighted averages of the changes (from the existing model to the new one) in separating the risk score distributions between events and non-events. However, ΔAP assigns heavier weights to the changes in higher-risk regions, whereas ΔAUC weights the changes equally. Due to this difference, the two IncV metrics can disagree, and the numerical study shows that their disagreement becomes more pronounced as the event rate decreases. In the numerical study, we also find that ΔAP has a wide range, from negative to positive, but the range of ΔAUC is much smaller. In addition, ΔAP and ΔsBrS are highly consistent, but ΔAUC is negatively correlated with ΔsBrS and ΔAP when the event rate is low.ConclusionsΔAUC treats the wins and losses of a new risk model equally across different risk regions. When neither the existing or new model is the true model, this equality could attenuate a superior performance of the new model for a sub-region. In contrast, ΔAP accentuates the change in the prediction accuracy for higher-risk regions.

Project description:Rationale and objectivesThe "binormal" model is the most frequently used tool for parametric receiver operating characteristic (ROC) analysis. The binormal ROC curves can have "improper" (non-concave) shapes that are unrealistic in many practical applications, and several tools (eg, PROPROC) have been developed to address this problem. However, due to the general robustness of binormal ROCs, the improperness of the fitted curves might carry little consequence for inferences about global summary indices, such as the area under the ROC curve (AUC). In this work, we investigate the effect of severe improperness of fitted binormal ROC curves on the reliability of AUC estimates when the data arise from an actually proper curve.Materials and methodsWe designed theoretically proper ROC scenarios that induce severely improper shape of fitted binormal curves in the presence of well-distributed empirical ROC points. The binormal curves were fitted using maximum likelihood approach. Using simulations, we estimated the frequency of severely improper fitted curves, bias of the estimated AUC, and coverage of 95% confidence intervals (CIs). In Appendix S1, we provide additional information on percentiles of the distribution of AUC estimates and bias when estimating partial AUCs. We also compared the results to a reference standard provided by empirical estimates obtained from continuous data.ResultsWe observed up to 96% of severely improper curves depending on the scenario in question. The bias in the binormal AUC estimates was very small and the coverage of the CIs was close to nominal, whereas the estimates of partial AUC were biased upward in the high specificity range and downward in the low specificity range. Compared to a non-parametric approach, the binormal model led to slightly more variable AUC estimates, but at the same time to CIs with more appropriate coverage.ConclusionsThe improper shape of the fitted binormal curve, by itself, ie, in the presence of a sufficient number of well-distributed points, does not imply unreliable AUC-based inferences.

Project description:BackgroundRisk prediction models have been proposed for various diseases and are being improved as new predictors are identified. A major challenge is to determine whether the newly discovered predictors improve risk prediction. Decision curve analysis has been proposed as an alternative to the area under the curve and net reclassification index to evaluate the performance of prediction models in clinical scenarios. The decision curve computed using the net benefit can evaluate the predictive performance of risk models at a given or range of threshold probabilities. However, when the decision curves for 2 competing models cross in the range of interest, it is difficult to identify the best model as there is no readily available summary measure for evaluating the predictive performance. The key deterrent for using simple measures such as the area under the net benefit curve is the assumption that the threshold probabilities are uniformly distributed among patients.MethodsWe propose a novel measure for performing decision curve analysis. The approach estimates the distribution of threshold probabilities without the need of additional data. Using the estimated distribution of threshold probabilities, the weighted area under the net benefit curve serves as the summary measure to compare risk prediction models in a range of interest.ResultsWe compared 3 different approaches, the standard method, the area under the net benefit curve, and the weighted area under the net benefit curve. Type 1 error and power comparisons demonstrate that the weighted area under the net benefit curve has higher power compared to the other methods. Several simulation studies are presented to demonstrate the improvement in model comparison using the weighted area under the net benefit curve compared to the standard method.ConclusionsThe proposed measure improves decision curve analysis by using the weighted area under the curve and thereby improves the power of the decision curve analysis to compare risk prediction models in a clinical scenario.

Dataset Information

Area under precision-recall curves for weighted and unweighted data.

Publications

Area under precision-recall curves for weighted and unweighted data.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets