Graphical Models

Bayesian Networks

These are directed graphs, where there is directed edge between two variables which shows cause-effect relation.

These are modeled using the conditional probability mostly.

Conditional Parameterization of Bayesian Graph

Markov Networks

These are undirected graphs. Which doesn't necessarily shows cause-effect but rather the connection between variables.

Using conditional probability in case of Undirected Graphical Models seems erratic because there is no direction and hence no natural conditioning.

In the case of Markov Models, we want to capture the affinity between connected random variables.