Nonparametric maximum likelihood estimation of a concave receiver operating characteristic curve via geometric programming
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
A receiver operating characteristic (ROC) curve plots the true positive rate of a classifier against its false positive rate, both of which are accuracy measures of the classifier. The ROC curve has several interesting geometrical properties, including concavity which is a necessary condition for a classifier to be optimal. In this paper, we study the nonparametric maximum likelihood estimator (NPMLE) of a concave ROC curve and its modification to reduce bias. We characterize the NPMLE as a solution to a geometric programming,a special type of a mathematical optimization problem. We find that the NPMLE is close to the convex hull of the empirical ROC curve and,thus, has smaller variance but positive bias at a given false positive rate. To reduce the bias, we propose a modification of the NPMLE which minimizes the L_1 distance from the empirical ROC curve. We numerically compare the finite sample performance of three estimators, the empirical ROC curve, the NMPLE, and the modified NPMLE. Finally, we apply the estimators to estimating the optimal ROC curve of the variance-threshold classifier to segment a low depth of field image and to finding a diagnostic tool with multiple tests for detection of hemophilia A carrier.
Abstract
A receiver operating characteristic (ROC) curve plots the true positive rate of a classifier against its false positive rate, both of which are accuracy measures of the classifier. The ROC curve has several interesting geometrical properties, including concavity which is a necessary condition for a classifier to be optimal. In this paper, we study the nonparametric maximum likelihood estimator (NPMLE) of a concave ROC curve and its modification to reduce bias. We characterize the NPMLE as a solution to a geometric programming,a special type of a mathematical optimization problem. We find that the NPMLE is close to the convex hull of the empirical ROC curve and,thus, has smaller variance but positive bias at a given false positive rate. To reduce the bias, we propose a modification of the NPMLE which minimizes the L_1 distance from the empirical ROC curve. We numerically compare the finite sample performance of three estimators, the empirical ROC curve, the NMPLE, and the modified NPMLE. Finally, we apply the estimators to estimating the optimal ROC curve of the variance-threshold classifier to segment a low depth of field image and to finding a diagnostic tool with multiple tests for detection of hemophilia A carrier.
참고문헌 (25)
- S. Boyd, "A tutorial on geometric program-ming", Optim. Eng, 8(1), pp.67~127 (2007)
- S. Boyd, "Convex Optimization" (2004), Cambridge University Press
- G. Campbell, "An application of Lomax distributions in receiver operating characteristic (ROC) curve analysis", Communication in Statistics, 22, pp.1681~1697 (1993)
- D. D. Dorfman, "Proper receiver operating characteristic analysis: The bigamma model", Academic Radiology, 4, pp.138~140 (1996)
- C. Feltz, "Maximum likelihood estimation of the survival functions of N stochastically ordered random variables", J. Amer. Statist. Assoc, 80(392), pp.1012~1019 (1985)
- P. A. Flach, "Repairing concavities in ROC curves", Proceedings of 2003 UK Workshop on Computational Intelligence, pp.38~44 (2003)
- "CVX: Matlab software for disciplined convex program-ming"[URL]
- S. Johansen, "The product limit estimator as maximum likelihood estimator", Scand. J. Statist, 5(4), pp.195~199 (1978)
- R. A. Johnson, "Applied Multivariate Statistical Analysis, Third edition" (1992), Prentice Hall, Inc
- E. Kaplan, "Nonparametric estimation from incomplete observations", J. Amer. Statist. Assoc, 53, pp.457~481 (1958)
- J. Lim, "Estimation of stochastically ordered survival func-tions via geometric programming", Journal of Compuational and Graphical Statistics
- J. Lim, "Matimum likelihood estimation of ordered multinomial parameters by geometric programing", Compuational Statistics and Data Analysis, 53, pp.889~893 (2009)
- C. Llyod, "Regression models for convex ROC curves", Biometrics, 56, pp.862~867 (2000)
- C. Llyod, "Estimation of a convex ROC curve", Statist. Probab. Lett, 59(1), pp.99~111 (2002)
- "Yet Another LMI Parser, Version 2.4"[URL]
- M. W. McIntosh, "Combining several screening tests: optimality of the risk score", Biometrics, 58(3), pp.657~664 (2002)
- C. E. Metz, ""Proper" binormal ROC curves: theory and maximum-likelihood estimation", J. Math. Psych, 43(1), pp.1~33 (1999)
- "ggplab: A Matlab toolbox for geometric programming"[URL]
- Y. Nesterov, "Interior-Point Polynomial Methods in Convex Pro- gramming, volume 13 of Studies in Applied Mathematics" (1994), SIAM Philadelphia
- J. Norcedal, "Numerical Optimization, Springer Series in Operations Re- search" (1999), Springer-Verlag
- X. Pan, "The "proper" binomial model: Parametric receiver operating characteristic curve estimation with degenerate data", Academic Radiology, 4, pp.380~389 (1997)
- M. S. Pepe, "The Statistical Evaluation of Medical Tests for Classification and Prediction" (2003), Oxford University Press
- F. Preparata, "Convex hulls of finite sets of points in two and three dimensions", Comm. ACM, 20(2), pp.87~93 (1977)
- J. Z. Wang, "Unsupervised multiresolution seg- mentation for images with low depth of field", IEEE Transaction on Pattern Analysis and Machine Intelligence, 23, pp.85~90 (2001)
- C. Yim, "Multiresolution 3-D range segmentation using focused cues", IEEE Transaction on Image Processing, 7, pp.1283~1299 (1998)