Multivariate predictive modeling and validation

Many problems in chemistry involve the prediction of one or more qualitative or quantitative properties based on the experimental data. Examples of such problems involve, for instance, the possibility of predicting protein or lipid content in food matrices based on NIR spectra or of diagnosing the onset of a disease through the MS or NMR analysis of serum samples. In the former case, the property to be predicted is of a quantitative nature, while in the latter, it is discrete (qualitative). This chapter presents the chemometric strategies most commonly used to formulate predictive models, i.e., models that relate one or more dependent variables Y (qualitative or quantitative) to a set of independent variables X.