An Iterative Improvement Approach for the Discretization of Numeric Attributes in Bayesian Classifiers

The Bayesianclassifier is a simple approachto classification that producesresults that are easy for people to interpret. In many cases, the Bayesianclassifieris at leastasaccurateas much more sophisticated learning algorithms that produceresultsthat are more difficult for people to interpret. To use numeric attributes with Bayesian classifier often requires the attribute values to be discretized into a number of intervals. We show that the discretization of numeric attributes is critical to successful applicationof the Bayesianclassifierandpropose a new method based on iterative improvement search. We compare this method to previous approachesand show that it resultsin significant reductionsin misclassificationerror and costson an industrialproblemof troubleshootingthe local loop in a telephonenetwork. The approachcan take prior knowledgeinto accountby improving upon a user-providedset of boundarypoints, or canoperateautonomously.

• Type: Conference
• Year: 1995
• Where: KDD
• Authors: Michael J. Pazzani
• Keywords: Data Mining, KDD 1995, Numeric Attributes, Simple Approachto Classification, Sophisticated Learning Algorithms