Extracting Decision Trees from Diagnostic Bayesian Networks to Guide Test Selection

Abstract


 
 
 In this paper, we present a comparison of five different approaches to extracting decision trees from diagnostic Bayesian nets, including an approach based on the dependency structure of the network itself. With this approach, attributes used in branching the decision tree are selected by a weighted information gain metric computed based upon an associated D-matrix. Using these trees, tests are recommended for setting evidence within the diagnostic Bayesian nets for use in a PHM application. We hypothesized that this approach would yield effective decision trees and test selection and greatly reduce the amount of evidence required for obtaining accurate classification with the associated Bayesian networks. The approach is compared against three alternatives to creating decision trees from probabilistic networks such as ID3 using a dataset forward sampled from the network, KL-divergence, and maximum expected utility. In addition, the effects of using χ2 statistics and probability measures for pre-pruning are examined. The results of our comparison indicate that our approach provides compact decision trees that lead to high accuracy classification with the Bayesian networks when compared to trees of similar size generated by the other methods, thus supporting our hypothesis.