Multivariate filters combined with interval partial least square method: A strategy for optimizing PLS models developed with near infrared data of multicomponent solutions

Near infrared spectroscopy (NIR) is a technique capable of rapidly generating rich chemical information. However, many chemical problems are limited to the low sensitivity and selectivity due to the spectral similarity of the components in the sample. Therefore, this study aimed to evaluate the use of multivariate filters combined with variable selection to optimize analytical parameters of partial least square (PLS) models developed with NIR data. This strategy was applied to 64 spectra of solutions containing ethanol, acetic acid, and lactic acid in 1-octanol. The multivariate filters evaluated were orthogonal signal correction (OSC), generalized least squares weighting (GLSW) and external parameter orthogonalization (EPO). Firstly, the multivariate filters were evaluated using the complete spectra and then with the variables selected by the interval partial least square (iPLS) algorithm. The figures of merit, such as accuracy, precision, linearity, sensitivity, and selectivity were used to evaluate the performance of the models. The PLS models for ethanol, acetic acid and lactic acid prediction showed a reduction of, respectively, 46%, 32% and 74% of RMSEP values after the use of multivariate filters combined with iPLS. The proposed strategy increased the analytical sensitivity and selectivity by up to 25 and 17 times, respectively. Therefore, the use of multivariate filters combined with the selection of variables by iPLS can make PLS models more sensitive, selective, and accurate for NIR data of multicomponent organic solutions.