An entropy-based discretization method for classification rules with inconsistency checking

Discretization is an effective technique in handling continuous attributes for data mining, especially for classification problem. Most entropy-based discretization methods are local and easy to lose valuable information in data. In this paper, we present a new entropy-based algorithm. Through an inconsistency checking, we may add/delete cut points on the basis of preliminary discretization scheme. So the interact on all attributes is taken into consideration in the discretization process which make our method possess a global property. Experiment indicates that with the same rule generator C4.5, our method can produce stronger rules than existing entropy-based discretization methods.