A chemometric analysis of dual pulse staircase voltammograms obtained in mixtures of ethanol, fructose and glucose

In a previous paper we reported that by using artificial neural networks (B NNs) to interpret data acquired using dual pulse staircase voltammetry (DPS V), it is possible to determine the concentrations of aliphatic compounds i n mixtures. This paper extends that work by probing the relationship betwee n the DPSV response and mixture composition by fitting the acquired data to a proposed model, using a method based on multiple linear regression (MLR) and analysis of variance (ANOVA). The results of this suggest that the sys tem is predominantly a linear function of the analyte concentrations, with additional contributions from a non-linear ethanol term and a fructose-etha nol interaction term. The common multivariate calibration techniques of MLR , principal component regression (PCR) and linear and non-linear partial le ast squares (PLS) were subsequently evaluated for the calibration of DPSV v oltammograms. These approaches are compared with the previously published a rtificial neural network (ANN) results. ANNs are found to give the lowest p rediction errors, although PLS and PCR are only slightly worse - this is as sumed to be due to their inability to model the fructose-ethanol interactio n and non-linearity of the ethanol response. (C) 2000 Elsevier Science S.A. All rights reserved.