|
Abstract:
|
In this paper, we present a methodology for creating a confidence region for a receiver operating characteristic (ROC) curve. This ROC curve is a measure of the ability of a classifier to distinguish between two groups. In our motivating example, we will refer to these groups as genuines and imposters. We first convert our ROC to polar coordinates for some center (c_x,c_y), a set of radii, R, and a set of angles, Q. We then bootstrap both the collection of imposters and the collection of genuines a large number of times. For each genuine and imposter replicate we calculate a new ROC, R*, in polar coordinates for the same set of angles, Q. Having done this we can find an ROC region that captures a specified proportion of the bootstrapped curves. We apply this methodology to data for classification of biometric identification systems, e.g. fingerprint recognition.
|
