Variable selection for kriging in computer experiments

An efficient variable selection technique for kriging in computer experiments is proposed. Kriging models are popularly used in the analysis of computer experiments. The conventional kriging models, the ordinary kriging, and universal kriging could lead to poor prediction performance because of their prespecified mean functions. Identifying an appropriate mean function for kriging is a critical issue. In this article, we develop a Bayesian variable-selection method for the mean function and the performance of the proposed method can be guaranteed by the convergence property of Gibbs sampler. A real-life application on piston design from the computer experiment literature is used to illustrate its implementation. The usefulness of the proposed method is demonstrated via the practical example and some simulative studies. It is shown that the proposed method compares favorably with the existing methods and performs satisfactorily in terms of several important measurements relevant to variable selection and prediction accuracy.