Simulation of multicriteria data

Abstract

For several reasons like benchmarking of MCDA methods, the MCDA community should be interested in the production of simulated multicriteria datasets based on real datasets. Our goal in this paper is to propose several methods to simulate these new multicriteria data from an existing dataset. Simulated data should be as similar as possible to the initial dataset, including the capture of specific structure in the initial dataset (if any). We propose here to study independent sample, PCA-based sample and copula-based sample and to determine which one best succeeds in generating new data on demand. The copula-based method seems the best one to reproduce specific links between criteria.$n$