Constructing ROC confidence bands based on functional ordering
Reliably estimating the variability of Receiver Operating Characteristic (ROC) curves is essential for the rigorous evaluation of diagnostic tests. This study introduces a novel nonparametric framework for constructing confidence bands for ROC curves by leveraging the tools of Functional Data Analysis (FDA). While traditional functional confidence regions often rely on centrality-based depth measures, our approach utilizes a "down-upward" ordering derived from the hypograph index. This method allows for a more flexible characterization of the curve distribution’s boundaries.
Applying this framework to bootstrap ROC trajectories, we evaluate its performance through an extensive set of simulations. Our results demonstrate that this ordering-based approach achieves competitive results compared to established methods in the literature.
Keywords: functional data; ROC curves; ordering