Adaptive Bayes Network Algorithm

Note: The Adaptive Bayes Network (ABN) algorithm is deprecated in Oracle Data Mining 11g. Oracle Data Miner does not support building ABN models; you can view ABN models created using the Oracle Data Mining 11g APIs.

Adaptive Bayes Network (ABN) is used to solve classification problems. ABN is an Oracle proprietary algorithm that provides a fast, scalable, non-parametric means of extracting predictive information from data with respect to a target attribute.

ABN selects the most important attributes and then uses them in the model build.

Note: ABN may not give good results when there is a small amount of source data. For example, Single Feature build, the default model type, results in rules; you do not usually get meaningful rules if the source data has fewer than 20,000 rows.

ABN predicts binary as well as multiclass targets.

This topic discusses

Single Feature Build ABN Model Type

This is the only ABN model type that produces rules. This model type builds a single multi-attribute feature.

The user specifies:

The algorithm determines:

You do not usually get meaningful rules when the source data has fewer than 20,000 rows.

Multi Feature Build ABN Model Type

This ABN model type does not generate rules. This model builds a pruned Naive Bayes model and one or more multi-attribute features.

The user specifies:

The algorithm determines:

Naive Bayes Build ABN Model Type

This ABN model type does not generate rules. This model builds a pruned Naive Bayes model.

The user specifies:

The algorithm determines:

How ABN Builds Models

ABN begins by ranking the attributes and then builds a Naive Bayes (NB) model as a baseline using fixed parameter settings (both thresholds set to 0) and the number of attributes (Naive Bayes Predictors) specified by the user taken in order from the ranked list. (This NB model is not exactly what you'd get by using the NB algorithm directly.) You may choose NB as the Model Type to stop the build process at this point. If you have run Attribute Importance to determine the number of attribute having positive influence on the predictive power of the model, then you can enter that number in Naive Bayes Predictors for the most efficient Naive Bayes model.

Next ABN begins to build a sequence of little "trees" called features. Each feature has a number of levels. The algorithm stops adding new levels when adding a new level doesn't add to the model's accuracy. When the depth set by this test is reached, a new feature is built with root node split on the attribute next in the ranked list of attributes.

At each step of the building process, the model is tested against the model prior to the last step, including the baseline NB model. In particular, when an individual feature completes building, it is tested versus the model without that feature, and if there's no improvement, the new feature is discarded (pruned). When the number of consecutive discarded features reaches a number set internally by the algorithm, ABN stops building and what remains is the completed model.