Development and validation of multivariate qualitative methods in the food field

Identificador:  TDX:1971

Handle: https://hdl.handle.net/20.500.11797/TDX1971

Autores:  López Vilardell, Maria Isabel

Resumen:
The main objective of this thesis is the development and validation of analytical methodologies focused on multivariate qualitative analysis. These methods are used to address problems that require a binary response (yes/no). In this way, the number of quantitative characterizations, which are usually more expensive, is reduced and decision-making is made quicker by the use of simple and low-cost analytical instrumentation. Multivariate qualitative methodologies can be used in a wide variety of analytical fields and, in this thesis; they have been applied to the food field (contamination or adulteration of foodstuff). The works carried out in this thesis tackle the different steps that cover multivariate qualitative analysis: (1) Data source: several spectroscopic techniques such as NIR and IR have been used and a Raman modality called Surface-Enhanced Raman Scattering spectroscopy (SERS) has been evaluated. Although SERS technique is similar to other implemented spectoscopies (NIR and IR), it can be still considered as an emergent technique in the analytical multivariate field. (2) Chemometric data treatment: several classification techniques based on their modelling power has been applied; well-known techniques (SIMCA e UNEQ), recent developed techniques (OCPLS) and a new technique developed in this thesis (partial least square density modeling, PLS-DM). Moreover, two different modelling strategies have been studied, one-class (untargeted) and multiclass (targeted). In the literature, most of the studies are based on the modelling of two or more categories (targeted) whereas few of them are based on the modelling of just one category (targeted). Although untargeted strategy is not new, it has been low exploit. (3) Validation: An attempt to clarify the terminology used in the qualitative method validation has been presented. A validation protocol has been proposed which include the use of Performance Characteristic Curves to estimate the performance parameters with quantitative connotation (concentration limits and unreliability region).