Estimation of partial area under multivariate ROC curve in the presence of measurement errors

Abstract

The data collection process, particularly in the medical field, often introduces measurement errors. These errors can have an impact on the outcomes and parameter estimates of the model. In the case of classification scenarios, the presence of measurement errors also affects the inherent cumulative and summary measures of the Receiver Operating Characteristic (ROC) curve. Despite being a crucial aspect of the ROC curve, only a limited number of researchers have explored the issue of measurement errors when estimating the partial area under their respective ROC curves (pAUC) within a univariate framework. This paper focuses on estimating the partial area under the multivariate ROC curve while considering the presence of measurement errors. A bias-corrected estimator is proposed and through simulations and real datasets, it is observed that using the proposed bias-corrected estimator it is possible to quantify the deviation from true accuracy. It is also observed that the proposed bias-corrected estimator has lesser Bias and minimum MSE when compared to the estimated area values.