| Within the category of multivariate calibration methods, the chemist can find a plethora of calibration methods (e.g. Principal Component Regression, Partial Least Squares, Classical Least Squares, Multiple Linear Regression, Ridge Regression, and Continuum Regression). The practitioner can find many different variants for each of these particular methodologies, leading to a multitude of approaches, which can be overwhelming. In spite of the extensive availability of multivariate calibration methods, two techniques, Principal Component Regression (PCR) and Partial Least Squares (PLS) (and many variants of these two) account for the majority of the papers published in the chemical literature. These two methods have been applied to an immense number of analytical scenarios producing very promising and often similar results. PLS is widely considered to produce better results than PCR in chemical applications, although the evidence for this is not compelling. The question of which approach produces the best results for a given application becomes more difficult to answer in complex multicomponent chemical systems. Despite the intrinsic difficulties involved in simulating complex chemical systems, an attempt has been made in this work to develop methods for simulating complex mixtures in order to study which multivariate calibration method perform the best, or at least under what circumstances one performs better than the other.;Although the work presented here is not definitively conclusive, it has shed light on what appear to be misconceptions on the difference between PLS and PCR in the literature; that is to say, there are no clearly defined differences in performance. This research represents for the first time that an attempt has been made to develop a statistical model for complex mixtures in the study of multivariate calibration. Now that this tool has been developed, it can be further refined and used to obtain insight on other tools for multivariate calibration in a similar fashion. |
