Risk factor analysis of foodborne pathogen infection using statistic and soft computing approaches
To develop appropriate prevention and control strategies for sporadic cases of illness, it is important to accurately model the system and analyze the risk factors. The objective of this study is to utilize both statistic and soft computing models to identify the significant risk factors for ' Salmonella' Typhimurium DT104 and non-DT104 infection in Canada, and compare the findings. Previous studies have focused on analyzing each risk factor separately using single variable analysis, or modelling multiple risk factors using statistic models, such as logistic regression models. In this study, both neural network models and statistic models are developed and compared to determine which method produces superior results. Genetic algorithms are further incorporated to extract the optimal subset of factors that provide an accurate classification. The genetic algorithm based neural classifier significantly outperform the statistic models and neural networks alone because either statistic models or neural networks alone are not able to consider factors' nonlinear interaction with maximum likelihood estimate, which selects the significant risk factor based on likelihood ratio test. A neuro-fuzzy based method for predicting 'Salmonella' Typhimurium infections is further proposed. In addition, neural network models are developed to study the effect of climatic factors for 'Salmonella' infections. Simulation studies show that neural networks perform better than corresponding linear, quadratic and cubic regression models in terms of correlation coefficients between 'Salmonella' infections and climate factors.