46-5 |
|
W. ZHANG and J. P. Norback. Dept. of Food Science, Univ. of Wisconsin, Madison, 1605 Linden Dr., 204 Babcock Hall, Madison, WI 53706-1565 It has been shown that certain shelf-stable pasteurized process cheese spreads could be a potential cause of foodborne botulism. Using the correct formulation for process cheese can control intrinsic environmental factors to prevent undesirable microorganism growth. Our objective is to provide decision-making tools to select safe process cheese formulation. Previous experiments to determine the outgrowth and toxin production of Clostridium botulinum in process cheese were conducted by Nobumasa Tanaka, Kathleen Glass and Eric Johnson. In this study, the decision tree was generated from these data by batch processing of training examples, which required parameter estimation, error rates evaluation and pruning for noisy data and overfitting. The formulation factors considered included moisture level, pH, cheese type, water activity, sodium chloride, sodium phosphate and other additives. The Max-Gain approach was used to choose formulation factors for prediction. The decision tree predictive model was evaluated by comparing it with artificial neural network models and statistical models built by logistic regression. With the decision tree generated, predictions for toxin production were made for untested formulations by extrapolating from available data sets. The attributes selected in order of importance by the Max-Gain approach provided the information to evaluate each formulation factor for its antibotulinal property. Graphical plots were made to assist with formulation adjustment, as well as to compare different model building approaches. The decision tree provided satisfactory accuracy to separate potential hazardous formulations from safe formulations. Based on these predictions, formulations can be adjusted to obtain safe pasteurized process cheese products. Logistic regression can then be used to estimate the probability for toxin production. These results suggest that the decision tree approach can be a convenient, efficient and reliable tool to predict microorganisms' response in terms of environmental conditions. It can be applied to assist manufacturers in developing safe process cheese formulations.
Session 46, Dairy Foods: General
|