Modeling and Analysis of Risk Factors for Salmonella Typhimurium DT104 and non-DT104 Infections: A Comparison of Logistic Regression and Neural Networks

To develop appropriate prevention and control strategies, it is important to accurately model and analyze risk factors for sporadic cases of diarrhoeal illness. Although traditional statistical methods are commonly used in risk factor studies, they can be limited in some respects. The objective of this study is to utilize intelligent models to identify significant risk factors for Salmonella (S.) Typhimurium DT104 and non-DT104 illness in Canada, and compare findings to those obtained with traditional statistical methods. Single variable analysis (SVA), Logistic regression models (LRs) and Feedforward error back-propagation artificial neural networks (FEBNNs) are used to classify DT104 and non-DT104 cases and controls and identif significant risk factors. Final results showed that the proposed FEBNNs have better results than the corresponding LRs in terms of predictive accuracy, errors and correlation.