Bayesian Networks

Abstract

AbstractMany practical applications require the specification of the joint probability density (JPD) of a large set of random variables. This task can be very difficult in practice. Bayesian and Markov networks are two graphically based models that make the specification of the JPD and the propagation of uncertainty simple and efficient because they exploit the conditional independence structure of the JPD. This entry deals with Bayesian networks, also known as Belief networks. Markov networks are discussed in another entry. Bayesian networks are defined and their usefulness is illustrated by examples of the two most commonly used Bayesian networks, namely, the multinomial and Gaussian networks. The questions about how Bayesian networks can be constructed, how Bayesian networks can be learnt from data, and how uncertainty can be propagated when new evidence becomes available are also discussed.