Applications of Partial Least Squares to Spectral Data for Quantitative Determinations of Chemicals in a Mixture

Abstract

The use of partial least squares (PLS) for the calibration of a spectrophotometer to determine the concentrations of chemicals in aqueous or solid samples is described. The method of partial least squares for spectral analyses is a good alternative to other conventional multivariate calibration methods such as multiple linear regression (MLR) and principal components regression (PCR). The similarities and differences between these methods are discussed. The predictive performance of PLS is compared with that of PCR on three sets of spectral data that exhibit either a severe collinearity problem and/or contain an excessive number of independent variables relative to the number of data points. The concept of cross-validation for choosing the optimal model is also discussed.