一种基于熵的连续属性离散化算法贺跃.pdfVIP

第 25 卷第 3期 计算机应用 V ol. 25 No. 3 2005年 3月 　 C ompu ter App lications 　 M ar. 2005 :100 1- 9081(2005)03 -0637 - 02 1 2 1 贺　跃, 郑建军 , 朱　蕾 (1. 北京理工大学信息科学技术学院, 北京 100081;2. 北京理工大学 管理与经济学院, 北京 100081) (zjj76983@bit. edu. cn) 　:连续属性离散化的 键在于合理确定离散化划分点的 个数和位置。为了提高无监督离 散化的效率, 给出一种基于熵的连续属性离散化方法。该方法利用连续属性的信息量(熵 )的特性, 通过对连续属性变量的自身划分, 最小化信息熵的减少和区间数, 并寻求熵的损失与适度的区间数之 间的最佳平衡, 以便得到优化的离散值。实验表明该算法是行之有效的。 :熵;连续属性;离散化;分类 :TP311. 13　:A A n entropy-based algorithm for d iscretization of continuou s variab les 1 2 1 HE Yue ZHENG ian-jun ZHU Lei 1. School of Information S cience and TechnologyBeijing Institu te of TechnologyBeijing 100081China 2. School of Managem ent and Econom icsB eijing Institute of TechnologyB eijing 100081Ch ina A b stractIt is very important to ascertain rationally the num ber and positions of split po ints for discretization of con tinuous variab les. T o mi prove the efficiency of unsuperv ised discretizationan en tropy-based algorithm w as propo sed for discretization of continuous variables. It m ade use of the characteristics of the in form ation conten tentropyof a con tinuous variab leand partitioned the continuou s variab le by itself for m inmi izing both the lo ss of entropy and the num ber of partitions in order to find the best balance bewt een the inform ation loss and a low number of partitionsso then ob tained an op tmi al discretization result. T he expermi ents show this app roach effective. K ey w ord sen tropycontinuous v ariab lediscretizationclassification 　　 , , 。 , (