关键词: 离散化, 数据挖掘, 类-属性相互依赖, 变精度粗糙集, 决策树

Abstract: Most data mining and induction learning methods can only deal with discrete attributes; therefore, discretization of continuous attributes is necessary. The author proposed a data discretization method based on statistical correlation coefficient. The method captured the interdependence between attributes and target class with the aim to select optimal cut points based on statistical correlation theory. In addition, the author incorporated Variable Precision Rough Set (VPRS) model to effectively control information loss. The proposed method was applied to breast tumor diagnosis and data of other fields. The experimental results show that this method significantly enhances the accuracy of classification of See5.

Key words: discretization, data mining, Class-Attribute Interdependence (CAI), Variable Precision Rough Set (VPRS), decision tree