Tree Augmented Naive BayesΒΆ

The Tree Augmented Naive (TAN) Bayes algorithm is based on the Chow-Liu algorithm.

The TAN algorithm is useful for constructing classification networks where a specific node of the model is target of the reasoning. The target node is used to construct a conditional Chow-Liu tree (i.e., a Chow-Liu tree over all nodes except the selected target) with the selected root as root of the tree. Weights are defined as conditional mutual information (conditional on the target), and all nodes (except target) have target as an extra parent.