Kurt Varmuza

1 Introduction Empirical, multivariate models ŷ = f(x) for calibration or classification are well established in chemometrics and other areas of data science. The performance of such models for predicting a numerical property y from a set of x-variables may heavily depend on the data pretreatment, particularly on the scaling of the x-variables. Widely used are scaling methods based on the spread of the x-variables, and usually the variables are processed separately but by the same method.