Near-infrared diffuse reflectance spectroscopy and neural networks for measuring nutritional parameters in chocolate samples.

A rapid and non-destructive method has been developed for the characterization of chocolate samples based on diffuse reflectance near-infrared Fourier transform spectroscopy (DRIFTS) and artificial neural networks (ANNs). This methodology provides a potentially useful alternative to time-consuming chemical methods of analysis. To assess its utility, 36 chocolate samples purchased from the Spanish market were analyzed for the determination of the main nutritional parameters like carbohydrates, fat, proteins, energetic value and cocoa content. Direct triplicate measurements of each sample were carried out by DRIFTS. Cluster hierarchical analysis was used for selecting calibration and validation data sets, resulting in a calibration set comprised of 19 samples and a validation data set of 17 samples. As it is common the presence of non-linear effect in reflectance spectroscopy, ANNs was chosen for data pretreatment. The root-mean-square error of prediction (RMSEP) values obtained for carbohydrates, fat, energetic value and cocoa were 1.0% (w/w), 1.0% (w/w), 50 kJ (100 g)(-1) and 1.4%, respectively. The mean difference (d(x-y)) and standard deviation of mean differences (s(x-y)) of the carbohydrates, fat, proteins content, energetic value and cocoa content were 0.9 and 2.4% (w/w), 0.2 and 1.0% (w/w), 9.1 and 50 kJ (100 g)(-1), and -0.5 and 1.4%, respectively. The maximum relative error for the prediction (QC) of any of these parameters for a new sample did not exceed 5.2%.