Abstrato

Kinetic-spectrophotometric determination of tin species using feed- forward neural network and radial basis function networks in water and juices of canned fruits

MaryamAbbasi-Tarighat

Feed- forward neural network (FFNN) and radial basis function networks (RBFN) were used in the development of a kinetic-spectrophotometric method for the simultaneous determination of Sn(II) and Sn(IV). The twoway data matrices, based on changes of absorbance at themaximumwavelength of reaction products of Sn(II) and Sn(IV) with pyrocatechol-violet in acetate buffered solution (pH 4.0) were processed separately by the principal component-radial basis function-artificial neural network (exact fit and fewer neurons) and principal component feed-forward neural network (PC-FFNN). The network architecture (number of hidden, and output nodes), transfer functions, number of epochs, momentum and learning rate in FFNN model and spread value in radial basis function, were also optimized for getting satisfactory results with minimum errors. The proposed methods were successfully applied to determination of desirable metal ions in several synthetic samples. The results obtained by PC-FFNN and PC-RBF networkswere compared to each other. The prediction performance of RBF network (exact-fit) was better than RBF (fewer neurons) network and PC-FFNN. The obtained satisfactory results indicate the applicability of ANNs approach for determination of desirable species. The proposed methods were successfully applied to the quantification of the Sn(IV) and Sn(II) in different water samples and canned products.