A factor breakdown can be carried out according to various different algorithms. In near-IR spectroscopy, preference has been given to factor breakdown according to principal components (principal component analysis), or PLS factors. Factor breakdown according to principal components (PCA) makes use exclusively, for the determination of the factors, of the spectral information from the samples, and can therefore be carried out even before further findings regarding the samples are available. As already indicated, the factors are the newly-defined variables with which work is continued. In addition to the factors, what are referred to as the loadings are also determined in the principal component analysis; these describe how the factors interact with the original wavelengths.
|PCA and PLS for the factor breakdown|
In the factor breakdown according to PLS factors, in order to determine the factors it is not only the spectral variance which is taken into account , but also the variance of the content substances.
|PCA factor breakdown with n samples, m wavelengths in the spectrum, and f factors|
As a rule, this leads to the strength of significance of the factors for the analytical problem being increased, because the factors are oriented in accordance with the specified content substances.
|PLS factor breakdown with n samples, m wavelengths in the spectrum, i content substances, and f factors.|
The factors are presented during the calculation in such a way that the variation of the content substances can be explained as well as possible by the PLS factors which are to be determined. PLS1 and PLS2 factors differ in that, with PLS1 only one content substance is taken into account in the factor breakdown, while with PLS2 several content substances are considered.
Factors will be correlated with specific features of the samples, because the variation in the NIR spectra is caused by sample properties and characteristics. This can be observed in a comparison of score values (factor characterisation of a sample) and content substance volumes.