A Bayesian Discretizer for Real-Valued Attributes

Discretization of real-valued attributes into nominal intervals has been an important area for symbolic induction systems because many real world classiication tasks involve both symbolic and numerical attributes. Among various supervised and unsupervised discretization methods, the information gain based methods have been widely used and cited. This paper designs a new discretization method, called the Bayesian discretizer, and compares its performance with some other methods including the information gain methods implemented in C4.5 and HCV (Version 2.0). Over the 7 tested datasets, the Bayesian dis-cretizer has the best results for 3 in terms of predictive accuracy.