Application of artificial neural networks for predicting the thermal inactivation of bacteria: a combined effect of temperature, pH and water activity

An artificial neural networks (ANNs) prediction is proposed based on effects of temperature (T), pH and water activity (a(w)) on the thermal inactivation of Escherichia coli. Using the data of Reichart [Reichart, O. (1994). Modelling the destruction of Listeria monocytogenes on the base (sic) of reaction kinetics. International Journal of Food Microbiology, 23, 449-465)], root-mean-squares error and determination coefficient (R-2) Of prediction were 0.144 and 0.949, 0.232 and 0.868 and 0.234 and 0.815 for ANN, response surface methodology and Cerf's model [Cerf, O., Davey, K.R. & Sadoudi. A.K. (1996). Thermal inactivation of bacteria - a new predictive model for the combined effect of three environmental factors: temperature, pH and water activity. Food Research International, 29, 219-226.], respectively. This result demonstrates higher accuracy of ANN models than those of other two models. Of these three environmental factors, temperature was most important for thermal inactivation of E. coli. The other two factors were less but equally important. The effects of thermal inactivation of bacteria were neither linear nor second-order form of T, pH and a,, instead being in more complicated nonlinear form. The superiority of the ANN-based approach is due to high prediction accuracy and the ability to compute combined effects of environmental factors on the thermal inactivation rate of the bacteria. (C) 2001 Elsevier Science Ltd. All rights reserved.