ABSTRACT
In the previous chapters various types of designs for computer experiments were introduced. Once the data have been collected from an experiment, we wish to find a metamodel which describes empirical relationships between the inputs and outputs. Notice that the outputs of computer experiments are deterministic (i.e., no random errors); therefore, we describe the relationship between the input variables and the output variable by the model (1.1) or
output variable = f(input variables), (5.1)
where f is an unspecified smooth function to be approximated. Note that compared to nonparametric regression models in the statistical literature, (5.1) does not have a random error term on the right-hand side. In this chapter, we aim to extend regression techniques existing in statistical literature to model computer experiments. Let us begin with some fundamental concepts in statistical modeling.
