Artificial neural networks and partial least squares regression for pseudo-first-order with respect to the reagent multicomponent kinetic-spectrophotometric determinations

Abstract

Partial least squares (PLS) regression and an artificial neural network (ANN) were tested as calibration procedures for the kinetic-spectrophotometric determination of binary mixtures when the concentration of the reagent is much lower than the concentration of the analytes. The two calibration methods were first applied to computer-simulated kinetic-spectrophotometric data. The spectra of the reaction products (P₁, P₂) were represented by Gaussian bands with the same bandwidth and the effect of spectral overlap and experimental noise was studied. If both spectra are identical, the mixture cannot be resolved. However, if they are not, then it is possible to quantify simultaneously both analytes by measuring the absorbance at several wavelengths and times. It was found that the precision of the results depends fundamentally on the noise level in the rate constants. Both mathematical procedures were applied to the determination of benzylamine–butylamine mixtures using salicylaldehyde as chromogenic reagent. ANNs outperformed the PLS results giving a relative standard error of prediction of about 4% for the whole set of mixtures.