COMPUTATIONAL NEURAL NETWORKS IN CONJUNCTION WITH PRINCIPAL COMPONENTANALYSIS FOR RESOLVING HIGHLY NONLINEAR KINETICS

A method based on the use of an orthogonal linear filter, principal co mponent analysis (PCA), for preprocessing data used as input for a fee d-forward neural network is proposed. The method analyzes the signific ance of the eigenvalues of the correlation matrix associated with the first principal components of the data in order to select the subset o f principal components for the sample that provides the optimum genera lization value. The generalization error was estimated by using the le ave-one-out method, because it provides the most reliable results for the fairly small data set used. The performance of the proposed method was assessed by applying it to the resolution of mixtures of species exhibiting a very similar kinetic behavior in the presence of a mutual kinetic (synergistic) effect. In addition, use of the continuous-addi tion-of-reagent (CAR) technique, a second-order approach, increased th e nonlinearity of the system studied. Based on the results, the propos ed designs provide accurate estimates in the kinetic resolution of bin ary mixtures, with errors of prediction about 5%. The results obtained in this respect are quite good taking into account that the kinetic b ehavior of the mixtures studied conforms to highly complex differentia l equations.