Oracle Adaptive Bayes
Adaptive Bayes Network (ABN) constructs Bayesian Network classifiers by using Minimum Description Length (MDL) and automatic feature selection. ABN does well in certain situations where Naive Bayes does poorly and does at least as well in most other situations, although performance may be slower. The ABN algorithm provides the ability to build three types of advanced, Bayesian-based models, including simplified decision tree (single-feature), pruned Naive Bayes, and boosted multifeature models.
Generated Models
In single-feature build mode, ABN produces a simplified decision tree, based on a set of human-readable rules, that allows the business user or analyst to understand the basis of the model's predictions and act or explain to others accordingly. This may be a significant advantage over Naive Bayes and multifeature models. These rules can be browsed like a standard rule set in IBM SPSS Modeler. A simple set of rules might look like this:
IF MARITAL_STATUS = "Married"
AND EDUCATION_NUM = "13-16"
THEN CHURN= "TRUE"
Confidence = .78, Support = 570 casesPruned Naive Bayes and multifeature models cannot be browsed in IBM SPSS Modeler.