Structure discovery in Bayesian networks: An analytical tool for analysing complex animal health data

Abstract

Analysing animal health data can be a complex task as the health status of individuals or groups of animals, might depend on many inter-related variables. The objective is to differentiate variables that are directly associated with health status and therefore promising targets for intervention, from variables that are indirectly associated with health status and can therefore at best only affect this indirectly through association with other variables. Bayesian network (BN) modelling is a machine learning technique for empirically identifying associations in complex and high dimensional data, so-called “structure discovery”. An introduction to structure discovery using BN modelling is presented, comprising the key assumptions required by the methodology, along with a discussion of advantages and limitations. To demonstrate the various steps required to apply BN structure discovery to animal health data, illustrative analyses of data collected during a previously published study concerned with exposure to bovine viral diarrhoea virus in beef cow-calf herds in Scotland are presented.