Some Aspects on the Utility of Distance Measures in Comparing Two MROC Curves

Abstract

Receiver Operating Characteristic (ROC) curve is a widely used and accepted tool to assess the performance of a classifier or procedure. Apart from this, comparing the procedures or ROC curves is also of interest. A multivariate extension of ROC (MROC) curve that considers a linear combination of several markers for classification was proposed by Sameera, Vishnu Vardhan and Sarma [13]. In the present paper, some inferential procedures are given to compare two MROC curves by means of distance measures based on scores of MROC curve and summary measures such as mean vectors and dispersion matrices. Real and Simulated data sets are used to demonstrate the above proposed inferential aspects.