Critical Aspects of Supervised Pattern Recognition Methods for Interpreting Compositional Data

A lot of multivariate data sets of interest to scientists are called compositional or "closed" data sets, and consists essentially of relative proportions. A recent search on the web by entering "chemical compositional data", led to more than 2,730,000 results within different fields and disciplines, but specially, agricultural and food sciences (August 2011 using Google searcher). The driving causes for the composition of foods lie on four factors (González, 2007): Genetic factor (genetic control and manipulation of original specimens), Environmental factor (soil, climate and symbiotic and parasite organisms), Agricultural factor (cultures, crop, irrigation, fertilizers and harvest practices) and Processing factor (post-harvest manipulation, preservation, additives, conversion to another food preparation and finished product). But the influences of these factors are hidden behind the analytical measurements and only can be inferred and uncover by using suitable chemometric procedures.