PREDICTION OF THE GROWTH BEHAVIOR OF AEROMONAS HYDROPHILA USING A NOVEL MODELING APPROACH: SUPPORT VECTOR MACHINE

A new technique called support vector machine (SVM), which is used to predict the microbial growth, is presented in this paper. Experimental data on temperature, pH and NaCl from a previously published paper were modeled as inputs, and the kinetic growth parameters, including generation time (GT) and lag phase duration (LPD), were used as outputs of the SVM model. The results derived from SVM model, published artificial neural network (ANN) model and a traditional statistical model were compared using several evaluation criteria: accuracy factor (A(f)), bias factor (B-f), mean relative percentage residual, mean absolute relative residual, root mean square residual, internal validation (Q(2)) and external validation (Q(ext)(2)). Graphical plots were also used for model comparison. The results show that SVM outperforms ANN and statistical model on predicting microbial GT and LPD, especially when predicting data not used for model development. Sensitivity analyses of the three environmental factors show that the most influential on LPD of Aeromonas hydrophila is temperature, followed by pH and NaCl, and the most influential on GT is pH, followed by temperature and then NaCl. Practical ApplicationsAeromonas hydrophila is a foodborne microorganism, which has been isolated from seafood, red meat, poultry, vegetables and water. Predictive microbiology describes microbial behavior using mathematical models to predict the microbiological safety and quality of a food. A nonlinear modeling technique, termed support vector machine (SVM), is proposed to predict lag phase duration and generation time of A.hydrophila. Statistical criteria show that SVM has a better predictive performance for the generation time and lag phase duration than neural network and traditional statistical models. Sensitivity analyses of the SVM can reveal which condition is the most influential factor on the lag phase duration and generation time, respectively. Therefore, SVM could become an alternative to other modeling approaches, all of which are faster than traditional microbiological experiments.