SELECTION OF VARIABLES BY GENETIC ALGORITHMS TO CLASSIFY APPLE BEVERAGES BY ARTIFICIAL NEURAL NETWORKS

• Published in: Applied artificial intelligence 2005-02, Vol.19 (2), p.181-198
• Main Authors: Gestal, Marcos, Gómez-Carracedo, María Paz, Andrade, Jose Manuel, Dorado, Julián, Fernández, Esther, Prada, Darío, Pazos, Alejandro
• Format: Article
• Language: English
• ISSN: 0883-9514; 1087-6545
• DOI: 10.1080/08839510590901921

The importance of fruit beverages, and of apple juice in particular, in daily food habits makes juice authentication an important issue in order to avoid fraudulent practices and to protect human health. Among the instrumental techniques available in analytical laboratories, infrared spectrometry (IR) is a fast and convenient technique to perform screening studies in order to assess the quantity of pure juice in commercial beverages. The information gathered from the IR analyses has some "fuzzy" characteristics (random noise, unclear chemical assignment, etc.) and, therefore, advanced computation techniques (Artificial Neural Networks or ANNs) are needed to develop ad hoc classification models. Disappointingly, the large number of variables derived from IR spectrometry makes ANNs require too much training time. As a result, this work studies two different approaches to apply genetic algorithms as a suitable method to select a small subset of variables intended to optimize the development of the ANN models. Their performance will be compared among them and with several linear methods as well.