NEURAL NETWORKS VS PRINCIPAL COMPONENT REGRESSION FOR PREDICTION OF WHEAT-FLOUR LOAF VOLUME IN BAKING TESTS

Neural networks (NN) provide a simple means of predicting outcomes tha t depend upon complex, possibly nonlinear, relationships between many variables. A trained neural network was created and used to predict lo af volume of breads made from different wheat cultivars. Although crea ting the NN required specialized skills and considerable computational time, using the ''trained'' NN to estimate remix loaf volume, was ver y rapid and required only basic computer skills. Random Centroid Optim ization (RCO) was also employed to choose the best training parameters : learning rate = 0.820, smoothing factor = 0.723, noise = 0.056, numb er of hidden neurons = 5. NN was more accurate, faster and easier than Principal Component Regression Analysis.