Artificial Neural Networks for Multicomponent Kinetic Determinations

Marcelo Blanco, Jordi Coello, Hortensia Iturriaga, Santiago Maspoch, Miguel Redón

Research output: Contribution to journalArticleResearchpeer-review

74 Citations (Scopus)


An artificial neural network (ANN) procedure that uses the scores of a principal component model as input data was tested for calibration in the resolution of binary mixtures from kinetic measurements. The results thus obtained are compared with those provided by partial least-squares (PLS) regression and principal component regression (PCR). The ANN was first applied to simulated single wavelength kinetic curves. The effect of experimental variability was considered by assuming rate constants to fit a normal distribution curve. An amount of instrumental noise was also added to the simulated curves. Both linear and nonlinear systems were tested. Non-linearity was assumed to result from interactions between analytes and modeled by introducing a multiplicative term in the rate equation. The results provided by the three methods on linear systems were comparable; in the presence of interactions between analytes, however, the ANN method clearly outperformed the other two. The ANN method was also used to resolve mixtures of Fe(II), Co(II), and Zn(II) by displacement from their EGTA complexes with 4-(2-pyridylazo)resorcinol (PAR) using a stopped-flow injection assembly including a diode array detector. Preliminary experiments revealed the Co(II) and Zn(II) displacement reactions to be pseudo-first-order and that of Fe(III) to be a multistep process that departed from the linear behavior of the other two. Again, the results obtained with ANN were better than those provided by PCR and PLS. © 1995, American Chemical Society. All rights reserved.
Original languageEnglish
Pages (from-to)4477-4483
JournalAnalytical Chemistry
Issue number24
Publication statusPublished - 1 Jan 1995


Dive into the research topics of 'Artificial Neural Networks for Multicomponent Kinetic Determinations'. Together they form a unique fingerprint.

Cite this