We address the problem of outliers detection in a binary out-ranking relation. These elements are supposed to be rare, dissimilar to the majority of other elements and are likely to influence the outcomes of the considered method. We propose a model based on the distance introduced by De Smet and Montano and extend it to different samplings of the set of alternatives (which are used as a comparison basis). This leads to study the distribution of distance values. The presence of outliers is detected by the identification of bi-modal distributions. We illustrate this on examples based on the Human Development Index, the Environmental Performance Index (where artificial outliers are added) and the Shanghai Ranking of World Universities.
展开▼