Modeling Data from Computer Experiments: An Empirical Comparison of Kriging with MARS and Projection Pursuit Regression

Computer experiments enable scientists to study complex processes by running computer codes that simulate them. We consider here the analysis of data from computer experiments, comparing three methods for non-parametric smoothing of high-dimensional data: Kriging, Multivariate Adaptive Regression Splines (MARS) and Projection Pursuit Regression (PPR). On several data sets of varying complexity, we find that Kriging is consistently the best performer.