A Rank Graduation Accuracy Measure

Abstract

A key point in the application of data science models is the evaluation of their accuracy. Statistics and machine learning have provided, over the years, a number of summary measures aimed at measuring the accuracy of a model in terms of its predictions, such as the Area under the ROC curve and the Somers’ coefficient. Our aim is to present an alternative measure, based on the distance between the predicted and the observed ranks of the response variable, which can improve model accuracy in challenging real world applications.

Keywords: Predictive accuracy, Concordance measures, Credit Scoring