Evaluation of model parameter accuracy by using joint confidence regions: application to low complexity neural networks to describe enzyme inactivation

An existing low complexity, black box artificial neural network model (ANN model) is investigated towards its more general applicability in the field of high isobaric-isothermal inactivation of enzymes. The use of this non-li near modeling technique makes it possible to describe accurately synergisti c effects of pressure and temperature in contrast with more classical model s used in this novel area of food processing. The modeling approach will be illustrated on a new experimental data set, b eing used to validate the structural characteristics of the selected ANN mo del. Moreover, joint confidence regions, taking into account the correlatio n between model parameters, will be constructed. The results will be transl ated towards the raw experimental data. (C) 1998 IMACS/Elsevier Science B.V .