Development of a Secondary Model for the Growth of Salmonella enterica in Food by Applying Artificial Neural Networks and Databases (ComBase and FoodData Central)

Abstract

The secondary growth model for Salmonella was developed based on the artificial neural network (ANN) with data collected from ComBase and FoodData Central. In addition to the existing secondary model variables (temperature, pH, Na+, and water contents), more input variables (sugar, carbohydrate, lipid, and protein contents) were considered. The output variables were microbial growth parameters (lag phase duration [l] and maximum growth rate [mmax ]). A commercial ANN program (NeuralWorks Predict) was utilized with training at 80%, validation at 10%, and test data at 10%. ANN models were created using all data and cleansed data. Using the cleansed data, the training/testing root mean square error (RMSE) for mmax improved from 0.14/0.16 to 0.11/0.14, whereas the RMSE for l was still not acceptable, from 11.94/33.03 to 7.09/4.18. The l data were divided into two ranges with high and low goodness of fit, whereas the ANN model for each f ield was built, resulting in an optimally low RMSE.