Tangled Webs: Using Bayesian Networks in the Fight Against Infection

Mary Waterhouse1,2 and Sandra Johnson1

1Queensland University of Technology, Brisbane, Australia

2Wesley Research Institute, Brisbane, Australia

20.1 Introduction to Bayesian Network Modelling

A Bayesian network (BN) is a probabilistic graphical model showing factors and their interactions, relating to a response of interest (Jensen and Nielsen 2007; Pearl 1988). The graphical model represents factors as nodes, and relationships between factors are shown as directed links which indicate the direction of the dependence between the nodes to form a directed acyclic graph (DAG) (Lauritzen and Sheehan 2003). An example of a DAG is shown in Figure 20.1. The link going from node C4 to node C2 indicates that node C4 is a parent of node C2. Similarly, node C2 is said to be a child of node C4. Apart from the parent–child relationship between two connected nodes, it is possible to distinguish three types of relationships between any three connected nodes in a BN (Korb and Nicholson 2010). These relationship types are serial, converging and diverging and are illustrated in Figure 20.1. The BN characteristics (d-separation and Markov property) inherent in these three relationships are important in their inferencing capabilities and simplification of probability calculations (Koller and Friedman 2009; Korb and Nicholson 2010). The relationships do not need to be causal in nature; however, representing causality in the BN ...

Get Case Studies in Bayesian Statistical Modelling and Analysis now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.