Comparing artificial neural network and statistical methods for estimating the survival of Escherichia coli O157:H7 in dry fermented sausage

The Canadian Food Inspection Agency (CFIA) requires the meat industry to ensure Escherichia coli (E. coli) O157:H7 does not survive in dry fermented sausage (salami) because a series of food-borne illness outbreaks have resulted from the presence of this pathogenic bacterium. The industry is in need of an alternative technique like 'predictive modeling' for estimating bacterial viability because traditional microbiological enumeration is a time-consuming and laborious method. Testing the accuracy and speed of artificial neural networks (ANNs) for this purpose is a current trend in predictive microbiological research, especially for online processing in industries. Two experimental data sets, one on interactive effects of different levels of pH, water activity (Aw), the concentration of allyl isothyocyanate (AIT) at various time intervals and the second on interactive effects of Lactoferrin, ethylene-diamine-tetraacetic-acid and sodium bi-carbonate during sausage manufacture in reducing Escherichia coli O157:H7 were used to develop predictive models using General Regression Neural Network (GRNN) (a form of ANN) and a statistical linear polynomial regression technique. Both models were compared for their prediction error using various statistical indices. GRNN predictions for training and test data sets had fewer and less serious errors when compared with the statistical model predictions. GRNN models were far superior and considerably superior respectively, for training and test sets than the statistical model. Because it is simple, fast and quite accurate, the ANN model can be used for online processing by research and development departments or quality control sections of the meat processing industry to ensure product safety, and specifically for processing to eliminate Escherichia coli O15:H7 from dry fermented sausage.