.: Marcos Gestal :.     -----


Datos de la Publicación:

Mari Paz Gómez, Marcos Gestal, Julián Dorado, José Manuel Andrade
Título: Linking chemical knowledge and genetic algorithms using two populations and focused multimodal search
Revista: Chemometrics and intelligent laboratory systems
ISSN: 0169-7439
Volumen: 87
Páginas: 173-184
Editorial: ELSEVIER
Fecha Publicación: Junio 2007
Factor de Impacto: 2.063
doi: 10.1016/j.chemolab.2006.12.002
Categorías WoS: Automation & Control Systems - Cuartil: Q1 - Tercil: T1 - Posición 5 de 52
Computer Science, Artificial Intelligence - Cuartil: Q1 - Tercil: T1 - Posición 16 de 93
Instruments & Instrumentation - Cuartil: Q1 - Tercil: T1 - Posición 6 de 55
Mathematics, Interdisciplinary Applications - Cuartil: Q1 - Tercil: T1 - Posición 5 de 74
Statistics & Probability - Cuartil: Q1 - Tercil: T1 - Posición 8 de 91
Chemistry, Analytical - Cuartil: Q2 - Tercil: T2 - Posición 27 de 70
Citas Google Scholar: 2


A classification method should establish the class to which a given object belongs to. Since not all experimental variables, that were measured on the samples yield the same quality and quantity of information, some of them can deteriorate the performance of the classification. Here, four strategies to perform variable selection in mid-infrared spectral data using genetic algorithms, GAs, are presented: fixed search, pruned search, multimodal search by hybrid two-population GA, HTP-GA, and focused HTP-GA. The former two are relatively simple whereas the latter two are more complex. Focused HTP-GA in particular was designed to allow introduction of chemical information in the GA finess function and, thus, get mathematical solutions which are also chemically-driven and, therefore, simplify the chemical understanding of the models. This is the first application in the Analytical field where the GA algorithm is used to find out a solution with chemical meaning because current applications proceed with a human variable interpretacion after selection took place. Select variables are feed to current backpropagation artificial neural network (ANN) in order to perform classification. Since both the GAs and ANNs are stochastic, it is hard to ascertain the true importance of the type of GA being used and, therefore, a twofold approach is followed. First, the GAs are tested using a fixed ANN topology for classification so that the differences on the classification results will (mainly) depend on the GA-selected variables. Second, once the variables were selected the best classification model was searched for optimizing the classifying ANN to ascertain how good the final results of the overall GA-ANN models can be. It is worth noting that the four approaches can be easily tailored to favour either the reduction in the number of variables, the classification abilities of the overall model or the chemical understanding. The practical problem considered here is to develop a screening procedure to ascertain the amount of pure juice used to prepare commercial apple-based beverages.