An extension of PROMETHEE to hierarchical multicriteria clustering

Abstract

Multicriteria clustering can be seen as a hybridisation between ranking and sorting problematic. These methods are used to build totally or partially ordered groups of alternatives based on preference relations. In the context of totally ordered clustering, two hierarchical approaches (top-down and bottom-up) based on PROMETHEE II have been developed in this paper. These methods rely on the optimisation of the clustering structure (by maximising the intra-cluster homogeneity and the inter-clusters heterogeneity). A third approach is developed as a hybrid model that merges the information obtained by both previous models. A specific quality index has been introduced to be able to evaluate the method's outputs and to choose appropriately the desired number of clusters. The three procedures have been tested on several dataset (Shanghai Ranking of World Universities, Environmental Performance Index and CPU evaluations) and the results have been compared with P2Clust.