INCOMPLETE AHP PAIRWISE MATRIX RECONSTRUCTION USING A NEURAL NETWORK-BASED MODEL

摘要

The Analytic Hierarchy Process (AHP) is a decision making method that has been widely used to obtain a ranking of alternatives, based on diverse criteria, that are placed in a hierarchical structure. At each level of the structure, the comparison of criteria is carried out using pairwise matrices. The decision maker completes matrices according to its judgments and the preference alternative values are obtained by the calculus of principal eigenvectors. If these matrices have missing elements, they must be completed previous to the calculation of the principal eigenvector. Also, if the comparison judgments between pairs are inconsistent, some methods to improve the consistency must be used, so that the results are coherent. In this paper we present a method based on a Multi Layer Perceptron (MLP) neural network that complete the pairwise matrices and improve its consistence simultaneously.