This paper describes and compares different kinds of statistical methods proposed in the literature as suited for solving calibration problems with many variables. These are: principal component regression, partial least-squares, and ridge regression. The statistical techniques themselves do not provide robust results in the spirit of calibration equations which can last for long periods. A way of obtaining this property is by smoothing and differentiating the data. These techniques are considered, and it is shown how they fit into the treated description.
HolstH., Statistical Analysis of NIR Measurements (in Danish) (The Institute of Mathematical Statistics and Operations Research, Lyngby, 1988), Ph.D. Thesis No. 52.
2.
NæsT. and MartensH, Comm. Stat. Sim. Comput.14, 545 (1985).
3.
HeilandI. S., “Some Theoretical Considerations around PLS,” in Data-Approximation by PLS-Methods, MartensM. Ed. (Norwegian Computing Center, Oslo, 1987), pp. 25–34.
4.
HeilandI. S., On the Structure of Partial Least Squares Regression (Department of Mathematics and Statistics, Agricultural University of Norway, Oslo, 1986), Report No. 21.
5.
MarkH.NorrisK. and WilliamsP. C, Anal. Chem.61, 398 (1989).
