Optimization of Computational Neural Network for Its Application in the Prediction of Microbial Growth in Foods

The power of computational neural networks (CNN) for microbiological growth prediction was evaluated. The training set consisted of growth responses data from a combination of three strains of Salmonella in a laboratory medium as affected by pH level, sodium chloride concentration and storage temperature. The architecture of CNN was designed to contain three input parameters in the input layer and one output parameter in the output layer. For their optimization, algorithms were developed to prune the net connections, obtaining an improvement in the generalization and a decrease in the number of necessary patterns for the training. The standard error of prediction (%SEP) obtained was under 5% using twenty inputs to the net, and the result was significantly smaller than the one obtained using regression equations. Therefore, the usefulness of CNN for modeling microbial growth is appealing, and its improvement promises results that will be better than those obtained by other estimation methods up to now.