Outlier detection in bilinear calibration

The evolution program, aiming to derive a clean subset from the contaminated calibration data set, was introduced and its performance tested with aid of the Monte Carlo method. Although the newly established procedure can be applied to any given calibration approach, in the case given in this paper we demonstrated its usefulness for the two most common bilinear calibration methods, namely of principal components regression and partial least squares regression. The presented results demonstrate that our approach allows building bilinear models in the presence of the multiple multivariate outliers.