Analysis of Classification of Discretization Method
Xinping Chen · 2020
Discretization method is of great significance in machine learning and data mining. It can improve the prediction accuracy. Compared with continuous variable, discrete variable is suitable for more traditional algorithms. A good discretization algorithm must minimize the information loss in the process of discretization and generate the appropriate number of cut-points. In this paper, the past discrete classification methods are sorted out and new classification criteria are proposed. Traditional discretization methods such as Eq.width, Eq.Freq, 1R, ID3, D2, MDL and adaptive quantizer are classified and briefly described. In addition, a new algorithm called continuous U-tree is evaluated. Finally, the goal and emphasis of discretization algorithm are analyzed, and the future development trend of discretization method is prospected.