Naive Bayesian learner. The Naive Bayesian learner makes
the simplifying assumption that all the attributes are
independent.
Data Driven Bayesian learner. The complex Bayesian
learner as implemented by Hugin attempts to learn
the structure of the network by looking at the correlation
between the attributes. Once the structure has
been determined data can then be used to determine
the node probability tables. The strength of a correlation
required to trigger the joining of two nodes can
be adjusted.
Expert constructed Bayesian network. When expert knowledge
of a given domain is to be represented as a BN
the usual process is for the domain expert(s) and BN
expert(s) to jointly construct the BN. If suYcient data
are available then the NPTs can be directly learnt and
then adjusted if required. However, when there is
insuYcient data to learn the NPTs these must also be
obtained from the expert(s).
K-nearest neighbour. K-nearest neighbour learners use a
likeness approach to prediction. That is, they look at
the instances most like the test case and usually have
some voting method by which the prediction is