A Max-Relevance-Min-Divergence criterion for data discretization with applications on naive Bayes

Shihe Wang, Jianfeng Ren, Ruibin Bai, Yuan Yao, Xudong Jiang

Abstract: Highlights•We identify the problem of lacking generalization ability in previous discretization methods.•The proposed Max-Dependency-Min-Divergence criterion simultaneously maximizes the discriminant information and generalization capability.•The proposed more practical Max-Relevance-Min-Divergence discretization can optimally discretize each attribute.•The proposed method outperforms SOTA discretization methods on 45 machine-learning datasets.