Multivariate Calibration for Quantitative Analysis

The multivariate calibration methods can be applied to any analytical technique and the ease at which multiparameter signals such as the absorbance at several wavelengths can be obtained in practice has facilitated its preferential expansion among spectroscopic techniques. Multivariate calibration methods offer the reliability for the results of unknown samples. With multivariate calibration, provided that a large number of variables are used for the analytical signal, examining the residuals can expose whether a given sample is " different " from those used for calibration and the result that it provides is thus unreliable. Some slight deviations from linearity can be modeled with multivariate calibration methods at the expense of an increased complexity, such as by using principal component regression (PCR) or partial least squares regression (PLSR) with additional principal components. Complex nonlinear systems can be resolved by using nonlinear calibration methods including some PCR and PLSR variants, or intrinsically nonlinear methods such as artificial neural networks (ANNs). The primary purpose of using a regression technique is constructing models allowing the value of the dependent variable, Y, which is usually a concentration, to be predicted from experimental data (absorbance in our case) represented by the independent variable, X. When multiple linear regression (MLR) analysis is used to construct a predictive model based on signals from a multianalyzer such as wavelengths, as inputs and a property of interest such as the concentration of a component, as output, the method is referred to as inverse linear regression. © 2009 Elsevier Inc. All rights reserved.