Fixed-Width Confidence Bands for ROC Curves with a Generalized False Positive Interval

• Activity Number: 228
• Type: Contributed
• Date/Time: Monday, August 3, 2009 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Defense and National Security
• Author(s): Daniel Katz*+ and Jacob Boon and Michael Miller and Austin DiOrio
• Companies: Johns Hopkins University Applied Physics Laboratory and Johns Hopkins University Applied Physics Laboratory and Johns Hopkins University Applied Physics Laboratory and Johns Hopkins University Applied Physics Laboratory
• Keywords: ROC curves ; bootstrap ; confidence bands

Abstract

This poster extends research on fixed-width confidence bands for receiver operating characteristic (ROC) curves by empirically validating its application to a generalized false positive interval. To generate the fixed-width confidence bands, we employ a bootstrapping technique combined with the software tool PROPROC (Metz et. al.), which fits ROC curves using maximum-likelihood estimation of a "proper" binormal ROC curve model. Given any false positive interval, we also use bootstrapping to compute an estimated standard error of the partial area under the ROC curve. This allows for statistical comparisons of two or more ROC curves based on the areas underneath the curves. This approach, which has been validated using computer simulation, has been applied to performance evaluations of new screening technologies by the Transportation Security Administration.