Validating the simulation metamodel: Some practical approaches

Abstract

This paper stresses the usefulness of developing a metamodel as an auxilliary model in simulation analysis and emphasizes the importance of validating the metamodel in order to determine whether it accurately approximates the simulation-generated data. A simulation model of the M/M/s queuing system was used for demonstration purposes. Two statistical validation procedures appropriate for simulation metamodels were discussed and demonstrated: use of a holdout sample and double cross- validation. The metamodel developed from the M/M/s queuing system simulation data proved to be valid as an approximation not only to the simulation but also to the real-world system itself.