Bayesian network analysis is a methodology for determining an optimal Bayesian Graphical model directly from observed data. The objective in a Bayesian network analysis is to derive a graph which best describes the interdependencies present in the data, for example between outcome variables of interest and other variables which might be risk factors. This is in contrast to other forms of Bayesian Graphical analyses which typically utilize a single graphical model whose structure is determined a priori using expert opinion. In such analyses the focus is not on determining the structure of the graph, but performing inference conditional on the chosen graph, for example to enable the estimation of some epidemiological parameter of importance such as the probability of disease transmission.

The case studies presented are ordered to provide a natural progression. We commence with two studies which utilize graphical models whose structure was derived a priori given expert knowledge of the epidemiological system under study. First, Greiner et al. apply graphical modelling to microbiological risk assessment, followed by David et al. whose focus is on a specific example of salmonellosis detection and source attribution using surveillance data. These two studies illustrate that a graphical model is simply a multidimensional probabilistic network which can be used to quantify the chance of particular events occurring given a set of events, actions or observations.

http://www.sciencedirect.com/science/article/pii/S016758771300038X

http://www.sciencedirect.com/science/journal/01675877/110/1