11.07.2015 Views

View - Universidad de Almería

View - Universidad de Almería

View - Universidad de Almería

SHOW MORE
SHOW LESS
  • No tags were found...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Proceedings of GO 2005, pp. 53 – 56.On weights estimation in MultipleCriteria Decision Analysis ∗R. Blanquero, 1 E. Carrizosa, 1 E. Con<strong>de</strong>, 1 and F. Messine 21 Facultad <strong>de</strong> Matemáticas, <strong>Universidad</strong> <strong>de</strong> Sevilla, Spain{rblanquero,ecarrizosa,educon}@us.es2 ENSEEIHT-IRIT, 2 rue C Camichel, 31071 Toulouse Ce<strong>de</strong>x, FranceFre<strong>de</strong>ric.Messine@n7.frAbstractKeywords:Several Multiple-Criteria Decision Making methods require, as starting point, weights measuringthe relative importance of the criteria. A common approach to obtain such weights is to <strong>de</strong>rive themfrom a pairwise comparison matrix A.There is a vast literature on proposals of mathematical-programming methods to infer weightsfrom A, such as the eigenvector method or the least (logarithmic) squares. Since distinct proceduresyield distinct results (weights) we pose the problem of <strong>de</strong>scribing the set of weights obtained by“sensible” methods: those which are Pareto-optimal for the nonconvex (vector-) optimization problemof simultaneous minimization of discrepancies.A characterization of the set of Pareto-optimal solutions is given. Moreover, although the abovementionedoptimization problems may be multimodal, standard Global Optimization methods cancome up with a globally optimal vector of weights in reasonable time.Pairwise comparison matrices, weights, Interval Analysis.1. IntroductionSeveral strategies have been suggested in the literature to associate with a set D = {d 1 , . . . , d n }of <strong>de</strong>cisions weights x 1 , x 2 , . . . , x n reflecting <strong>de</strong>cision-maker’s preferences. In the AnalyticHierarchy Process (AHP), [11, 12, 14, 15], an n × n matrix A,⎛⎞a 11 a 12 · · · a 1na 21 a 22 · · · a 2nA = ⎜⎝.. . ..⎟. ⎠a n1 a n2 · · · a nnis obtained after asking the <strong>de</strong>cision-maker (DM) to quantify the ratio of his/her preferencesof one <strong>de</strong>cision over another. In other words, for every pair of <strong>de</strong>cisions d i , d j , the term a ij > 0is requested satisfyinga ij ≈ x ix j(1)The matrix A so obtained must be a positive reciprocal matrix, i.e.,a ji = 1a ij> 0 for all i, j = 1, 2, . . . , n.∗ This research has been partially supported by grants BFM2002-04525-C02-02 and BFM2002-11282-E MCYT, Spain

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!