Although the consumption of fruit juices has increased sharply over the last years thanks to its unrivaled nutritional benefits and health-protection advantages, fraudulent fruit juices can sometimes be found in commercial facilities. Fortunately, new EU Directives stress the importance of labeling and food security and trazability. One of the most significant issues enforced by the new guides (already in force or to be published) is the requisite to state clearly the origin of the juice (whole fruit, directly extracted, concentrates or not-from-concentrate) as well as the total quantity of pure juice employed to prepare the beverage. Two cornerstones of the quality control of juice-based beverages are to monitor the amount of juice and the amount (and nature) of other substances added to the beverages. Particularly, sugar addition is a common and simple adulteration, though difficult to characterize. Other adulteration methods, either alone or combined, include addition of water, pulp wash, cheaper juices, colourants, and other undeclared additives (intended to mimic the compositional profiles of pure juices).
This work proposes Genetic Algorithms (GA) as a convenient alternative to select a small number of original variables (comprising the most important information of the original data sets) as a previous step before using ANNs to develop classification and/or prediction models. In order to compare the results, the number of variables is defined using a external criterion (here, only 2 variables).