An alternative characterization of a Bayesian network Original Research Article

To build a Bayesian network (BN), one may directly construct a directed acyclic graph (DAG) based on the causal relationships of the domain variables. However, it may be necessary in many applications to construct a DAG from the conditional independencies (CIs) supplied by different sources or experts. In this paper we provide an alternative characterization of the graphical structure of a BN. Based on this characterization, one can easily test whether an arbitrary set of CIs can be faithfully represented by a DAG.