Construction of stochastic simulation metamodels using smoothing splines

Metamodels are an analysis tool that better exposes the input‐output relationship of the simulation model. However, not much research has been done on nonparametric metamodels compared with parametric metamodels. Interpolating or smoothing splines are based on local fitting and can adapt to more complex shapes than traditional linear and nonlinear parametric metamodels. Additionally, the smoothing parameter of the smoothing splines can be used to control the effects of statistical noise in stochastic simulations. We suggest a spline metamodel construction methodology, based on a proposed experimental design that allows a selection of the smoothing factor that enhances the quality of the resulting metamodelling and minimises oscillation between design points.