PREDICTING MILK SHELF-LIFE BASED ON ARTIFICIAL NEURAL NETWORKS AND HEADSPACE GAS-CHROMATOGRAPHIC DATA

The usefulness of artificial neural networks (ANN) for milk shelf-life prediction by multivariate interpretation of gas chromatographic prof iles and flavor-related shelf-life was evaluated and compared to princ ipal components regression (PCR). The training set consisted of dynami c headspace gas chromatographic data collected during storage of paste urized milk (input information for the neural network used to make a d ecision) and its corresponding shelf-life (prediction or response). AN N had better predictability than PCR. A standard error of the estimate of 2 days in shelf-life resulting from regression analysis of experim ental vs predicted values indicated a high predictability of ANN.