A novel measure for evaluating an ordered list

Abstract

Sensitivity and specificity are the most widely used statistics for measuring the performance of a binary classification test. They stand vastly meaningful for variety of use cases where the classifying tests are affordable. But unfortunately, there is a legion of problems arriving from different streams of natural sciences where the screening test is too expensive to render for all the predicted objects. Thus, the trend has been for scientists to calculate the sensitivity and the specificity of a binary classification test based on a handful of experimentally proven facts, which is theoretically uncertain. In this article a novel measure is proposed that assigns importance to multiple ordered lists, taking into account the share of majority voted ranked pairs of elements a list contains. A real life bioinformatic application is demonstrated in the domain of microRNA target prediction where a number of algorithms exist. Using the proposed measure, we aim to assign certain weight to each algorithm that conveys its reliability with respect to the rest.