Abstract
Artificial neural networks (ANN) have widely been used in chemical analysis of mixtures. In this work, ANN was applied to determining concentrations of halide anions or oxoanions (chloride, bromide, iodide, acetate, carbonate and sulfate) which are very similar in property to each other, and so it is quite difficult to do that. In this case, absorption spectra appear in a shorter wavelength region. Therefore, we used a sensor molecule (a cobalt(III) complex) which shows charge-transfer bands owing to ion-pair formation with the anions. The charge transfer absorption upon addition of an anion is somewhat different from other ones. Multilayer neural networks were chosen for pattern-recognition analysis. The architecture consists of input, output, and hidden layers, which are where data are processed through a back propagation training algorithm. Of the 25 spectra obtained, 24 of these were used to train the data set (charge-transfer absorbance) by using 21 wavelengths from each spectrum. When the remaining one unknown was processed, an absolute error of a few % was obtained for the output concentrations for the mixtures containing chloride and bromide, bromide and iodide, bromide and acetate, and then carbonate and sulfate.
