论文信息 - A Relationship between the Average Precision and the Area Under the ROC Curve

A Relationship between the Average Precision and the Area Under the ROC Curve

For similar evaluation tasks, the area under the receiver operating characteristic curve (AUC) is often used by researchers in machine learning, whereas the average precision (AP) is used more often by the information retrieval community. We establish some results to explain why this is the case. Specifically, we show that, when both the AUC and the AP are rescaled to lie in [0,1], the AP is approximately the AUC times the initial precision of the system.

[1] David I. Laibson,et al. Golden Eggs and Hyperbolic Discounting , 1997 .

[2] P. Qiu. The Statistical Evaluation of Medical Tests for Classification and Prediction , 2005 .

[3] J. Hanley,et al. The meaning and use of the area under a receiver operating characteristic (ROC) curve. , 1982, Radiology.

[4] David J. Hand,et al. Measuring classifier performance: a coherent alternative to the area under the ROC curve , 2009, Machine Learning.

[5] Mark Goadrich,et al. The relationship between Precision-Recall and ROC curves , 2006, ICML.

[6] Stephen E. Robertson,et al. Modeling score distributions in information retrieval , 2011, Information Retrieval.

[7] Dale Schuurmans,et al. Augmenting Naive Bayes Classifiers with Statistical Language Models , 2004, Information Retrieval.