Multi-variable Discretization Based on Extended Maximum Information Coefficient

Supervised data discretization is partitioning continuous variables based on a label of classification. The label usually defaults to a categorical variable. But in some applying scenario, the label may be a continuous variable. The usual way to handle this problem is transferring the label to a categorical variable, then discretizing target variable. However, the error generated in the first step may be magnified in the second step. Therefore, we propose a data discretization method based on an extended maximum information coefficient which can deal with multiple variables to combine two steps in one. We analyze the mathematical properties and calculation strategy of the method, and give examples of data set from the world health organization.