Study about aggregation operators for the assessment of concordance level in outranking-based multi-criteria decision aiding methods

Identifier:  TFM:359

Handle: https://hdl.handle.net/20.500.11797/TFM359

Authors:  Orama, Ayebakuro Jonathan

Abstract:
In Multi-Criteria Decision Aiding (MCDA), aggregation operators are used to merge the information provided by different criteria about a set of alternatives. This overall score could then be used to construct a rank order or ordered classification of alternatives. The ELECTRE outranking method is inspired in voting procedures. It employs the Weighted Average operator (WA) to calculate the overall concordance index when constructing a pairwise outranking relation that evaluates the degree of truth of the assertion “option a is at least as good as option b”. The use of the weighted average operator has been observed to have an undesired compensative effect between opposite values. In this master thesis, we propose modifying the concordance calculation in ELECTRE by substituting the weighted average operator for operators from the Ordered Weighted Average (OWA) family of operators to give more flexibility to the definition of concordance as “a sufficient majority in favor of the outranking relation”. In particular, we study these three operators: OWAWA (Ordered Weighted Average Weighted Average), WOWA (Weighted Ordered Weighted Average), IOWA (Induced Ordered Weighted Average). They are appropriate operators because they are designed to combine the advantages of the Weighted Average operator and Ordered Weighted Average operator. That is, they retain the importance given to each criterion while adding also a conjunctive/disjunctive character to the definition of majority. The three proposed OWA-based outranking constructs are implemented as Web services for integration into the European Diviz workbench, so they are based on the ELECTRE concordance module already deployed in Diviz. Three different experiments were carried out with various parameterization of each proposed method, which were compared to the conventional method used in ELECTRE. These experiments show that OWAWA and WOWA are suitable to combine Weighted Averaging and Ordered Weighted Averaging. On the contrary, an undesired behaviour is observed in the IOWA operator which makes it unsuitable for use in ELECTRE.