Prediction of deterministic functions: an application of a Gaussian kriging model to a time series outlier problem

Abstract

Univariate kriging is used in approximating the sample distributions of statistics for detecting outliers in time series. Two experimental designs are compared, a traditional factorial design and a maximum entropy design. In both cases, the results, in terms of prediction errors, are satisfactory. The computing times for the predictors imply a reduction of two orders of magnitude in comparison to a Monte Carlo procedure.