Prediction from wrong models: the Kriging approach

This article reviews a very general, multi-factor, parametric approach to system modeling, called Kriging. It is shown how this method can lead to adequate prediction, together with an estimation of the accuracy of the predicted values of an observable, although the model is based on outrageously simplistic equations (and thus "wrong", from a phenomenological point of view). Kriging appears to be a useful approach for many applications in electromagnetics, in particular in those where computational time is an important issue. This is true when the simultaneous analysis of an observable with respect to more than one factor is needed.