Analysis of the behaviour of genetic algorithms when learning Bayesian network structure from data

Abstract In the last few years Bayesian networks have become a popular way of modelling probabilistic relationships among a set of variables for a given domain. For large domains, though, the construction of Bayesian networks is a hard task and the number of possible structures and the number of parameters for those structures can be huge. Trying to solve this, some researchers have studied how this construction can be automated. This work analyzes the behaviour of genetic algorithms when performing such automation. It is shown that the different ways in which genetic algorithms can tackle the problem influence the results.