A new approach to avoid rank reversal cases in the TOPSIS method. (June 2019)

• Record Type: Journal Article
• Authors: Aires, Renan Felinto de Farias; Ferreira, Luciano
• Is Part Of: Computers & industrial engineering. Volume 132(2019)
• Page Start: 84
• Page End: 97
• Publication Date: 2019-06
• Subjects: Multi-criteria decision-making -- TOPSIS -- Rank reversal -- Normalization; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2019.04.023
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: A new approach to solve the rank reversal problem in the TOPSIS method is presented. Simulated decision problems and a real case are used to validate the model. Classical normalization procedures are analyzed and compared. Different rank reversals cases are presented and analyzed. Abstract: During recent decades, different methods of Multicriteria Decision Support have been used to help decision-makers select better alternatives for various decision problems. However, these methods have been criticized in the literature because they present a problem called rank reversal. In particular, analyzing this problem in relation to the TOPSIS method is still limited. Our review of the literature showed that papers are limited to analyzing cases of rank reversal by adding and removing alternatives and the solutions proposed are limited to case studies and can be improved in order to widen the scope of their application. Thus, we initially performed an analysis to determine the main cases of the rank reversal presented in the literature and to identify the main gaps in relation to the TOPSIS method. Next, we define a framework for evaluating both the TOPSIS method and the proposed model in relation to different cases of rank reversal. Finally, this paper puts forward a new method called R-TOPSIS, which proved to be robust in the experiments performed, since there were no cases of rank reversal for either the simulated cases nor for the real case used to validate it. The proposed method was also validated using statistics of dispersion and similarity to evaluate its adherence to the classic TOPSIS method.