Artificial neural networks as a tool for soft-modelling in quantitative analytical chemistry: the prediction of the water content of cheese

The application of artificial neural networks for the modelling of a complex process was examined. A real data set concerning the batch production of cheese from an actual plant was used to predict the resulting water content of the cheese from the milk composition and process parameters. Owing to the complex nature of the data and the limited number of available patterns, difficulties were encountered when the standard backward error propagation algorithm was applied and no solution was derived. Several adaptions to the algorithm as suggested in the literature were then examined, and several gave satisfactory solutions. The resulting mean of the absolute values of the absolute prediction errors was 0.25% and 0.29% for known and unknown patterns, respectively, with a worst case error of 0.8%.