Half-global discretization algorithm based on rough set theory

作     者：Tan Xu Chen Yingwu 

作者机构：School of Information Systems & Management National Univ. of Defense Technology Changsha 410073 P. R. China 

出 版 物：《Journal of Systems Engineering and Electronics》 (系统工程与电子技术（英文版）)

年 卷 期：2009年第20卷第2期

页      面：339-347页

核心收录：

学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 0808[工学-电气工程] 0809[工学-电子科学与技术（可授工学、理学学位）] 081104[工学-模式识别与智能系统] 08[工学] 0839[工学-网络空间安全] 0802[工学-机械工程] 0835[工学-软件工程] 0811[工学-控制科学与工程] 081201[工学-计算机系统结构] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

主　　题：half-global discretization continuous condition attributes correlation coefficient rough entropy stability rough set theory 

摘      要：It is being widely studied how to extract knowledge from a decision table based on rough set theory. The novel problem is how to discretize a decision table having continuous attribute. In order to obtain more reasonable discretization results, a discretization algorithm is proposed, which arranges half-global discretization based on the correlational coefficient of each continuous attribute while considering the uniqueness of rough set theory. When choosing heuristic information, stability is combined with rough entropy. In terms of stability, the possibility of classifying objects belonging to certain sub-interval of a given attribute into neighbor sub-intervals is minimized. By doing this, rational discrete intervals can be determined. Rough entropy is employed to decide the optimal cut-points while guaranteeing the consistency of the decision table after discretization. Thought of this algorithm is elaborated through Iris data and then some experiments by comparing outcomes of four discritized datasets are also given, which are calculated by the proposed algorithm and four other typical algorithras for discritization respectively. After that, classification rules are deduced and summarized through rough set based classifiers. Results show that the proposed discretization algorithm is able to generate optimal classification accuracy while minimizing the number of discrete intervals. It displays superiority especially when dealing with a decision table having a large attribute number.