Modeling survival curves of Salmonella spp. in chorizos using artificial neural networks and regression

Time-dependent survival curves of Salmonella spp. in chorizos were modeled using both the classical method of statistical regression and the newly introduced method of artificial neural networks (ANNs). The survival curves were obtained experimentally for chorizos formulated at five initial water activity (A(w0)) levels of 0.85, 0.90, 0.93, 0.95 and 0.97, which had been stored under four different storage conditions: in a refrigerator (Ref) at 6C, at room temperature (RT) of 25C, in a hood (Hd) at 25C with forced inflow air circulation velocity F of 25.4 m/min and in an incubator (Inc) at 30C. The developed models enable prediction of survival curves (log count versus time) for Salmonella in chorizos as affected by a given set of operating conditions (A(w0), storage temperature [T] and F). Additionally, both the 1D- and 2D-values (time to reduce the count by 1 and 2 logs, respectively) were derived from a number of simulated survival curves and were used to develop regression models (R-2 = 0.980 and 0.977 for 1D- and 2D-value models, respectively) for predicting these two times as a function of operating conditions. Both 1D- and 2D-values increased with increasing A(w0) and decreasing T and F. Additionally, these times were more sensitive to A(w0) when the latter was above 0.940, and F was more influential at higher T. The ANN-based model (R-2 = 0.967) outperformed the regression-based model (R-2 = 0.919) and was also used to develop models for predicting the 1D- and 2D-values as a function of A(w0), T and F.