Data Discretization based on Maximum Information Coefficient

Most supervised discretization methods rely on category labels. However, in practical applications, supervisory information is not always category labels, it may also be continuous variables. Existing discretization methods cannot directly deal with the discretization, they need to convert it into category label, and then discretize other variables according to the category label. It is likely to amplify in the error in the second step because of the lack of feedback of the first step. In this paper, discretization method based on maximum information coefficient (MIC) is proposed. In the process of calculating MIC, the association between two continuous variables is measured by dividing data grids. Then the dividing intervals are given based on the division blocks. Because of the high time complexity of calculating MIC, this paper proposed a simplified method with equal-density intervals. Finally, the effectiveness of the method proposed in this paper is validated through case study.