Combination of Dichotomizers for Maximizing the Partial Area under the ROC Curve
In recent years, classifier combination has been of great interest for the pattern recognition community as a method to improve classification performance. The most part of combination rules are based on maximizing the accuracy and, only recently, the Area under the ROC curve (AUC) has been proposed as an alternative measure. However, there are several applications which focus only on particular regions of the ROC curve, i.e. the most relevant for the problem. In these cases, looking on a partial section of the AUC is the most suitable approach to adopt. In this paper we propose a new algorithm able to maximize only a part of the AUC by means of a linear combination of dichotomizers. Moreover, we empirically show that algorithms that maximize the AUC do not maximize the partial AUC, i.e., the two kinds of maximization are independent.
KeywordsClassifiers combination ROC curve partial AUC
- 2.Cortes, C., Mohri, M.: AUC optimization vs. error rate minimization advances. In: Neural Information Processing Systems. MIT Press, Cambridge (2003)Google Scholar
- 4.Jiang, Y., Metz, C.E., Nishikawa, R.M.: A receiver operating characteristic partial area index for highly sensitive diagnostic tests. Radiology 201, 745–750 (1996)Google Scholar
- 7.Hanley, J.A., McNeil, B.J.: The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143, 29–36 (1982)Google Scholar
- 9.Marrocco, C., Molinara, M., Tortorella, F.: Optimal linear combination of dichotomizers via auc. In: Proceed. of the 22nd Intern. Conf. on Mach. Learn. - Workshop on ROC Analysis in Mach. Learn., pp. 778–785 (2005)Google Scholar