Rank Correlation Coefficient Correction by Removing Worst Cases

Abstract

Rank correlation can be used to compare two linearly ordered rankings. If the rankings include noise values, the rank correlation coefficient will yield lower values than it actually should. In this paper, we propose an algorithm to remove pairs of values from rankings in order to increase Kendall's tau rank correlation coefficient. The problem itself is motivated from real data in bioinformatics context. © Springer-Verlag Berlin Heidelberg 2010.