An iterative improvement approach for the discretization of numeric attributes in Bayesian classifiers

摘要: The Bayesian classifier is a simple approach to classification that produces results are easy for people interpret. In many cases, the at least as accurate much more sophisticated learning algorithms produce difficult To use numeric attributes with often requires attribute values be discretized into number of intervals. We show discretization critical successful application and propose new method based on iterative improvement search. compare this previous approaches it in significant reductions misclassification error costs an industrial problem troubleshooting local loop telephone network. can take prior knowledge account by improving upon user-provided set boundary points, or operate autonomously.