| Abstract
| - The effects of resolution, spectral window, and backgroundtype on the predictive capability of partial least squaresregression (PLS) on spectra measured by an open-path Fouriertransform (OP/FT-IR) spectrometer were tested withspectra of mixtures of alkanes and chlorinated hydrocarbons.The results were compared with the results obtainedwith the identical data sets using classical least squaresregression (CLS). It is shown that the most accurate predictionsare obtained using the same conditions that wereoptimal for CLS, namely spectra measured at low resolutionand ratioed to background spectra over the same pathlength, with the calculations made over limited spectralwindows. However, good predictions could be achieved withbackground spectra measured over a very short path.Even in the worst cases, the relative error of predictionsmade by PLS was usually less than 5%. On average, thepredicted concentrations of the components of mixturescontaining up to five chemically similar analytes made usingthe PLS algorithm are 120 times more accurate than thepredicted concentrations of the components of the identicaldata sets made using CLS.
|
