Optimal Mean-Precision Classifier
For pattern recognition problems where a small set of relevant objects should be retrieved from a (very) large set of irrelevant objects, standard evaluation criteria are often insufficient. For these situations often the precision-recall curve is used. An often-employed scalar measure derived from this curve is the mean precision, that estimates the average precision over all values of the recall. This performance measure, however, is designed to be non-symmetric in the two classes and it appears not very simple to optimize. This paper presents a classifier that approximately maximizes the mean precision by a collection of simple linear classifiers.
KeywordsPattern recognition performance evaluation information retrieval precision-recall graph
Unable to display preview. Download preview PDF.
- [BS05]Brefeld, U., Scheffer, T.: AUC miximizing support vector learning. In: Proceedings of ICML 2005 workshop on ROC analysis in Machine Learning (2005)Google Scholar
- [FFHO02]Ferri, C., Flach, P., Hernandez-Orallo, J.: Learning decision trees using the area under the ROC curve. In: Proceedings of the ICML (2002)Google Scholar
- [Fla03]Flach, P.: The geometry of ROC space: understanding machine learning metrics through ROC isometrics. In: Proceedings of the international conference on Machine learning 2003, pp. 194–201 (2003)Google Scholar
- [NHBM98]Newman, D.J., Hettich, S., Blake, C.L., Merz, C.J.: UCI repository of machine learning databases (1998)Google Scholar
- [vR79]van Rijsbergen, C.J.: Information Retrieval, 2nd edn. Butterwort (1979)Google Scholar