一种连续条件属性值的决策表的归纳学习方法  

An Inductive Learning Approach of Decision Table with Continuous Condition Attribute Values

作　　者：李如强[1] 陈进[1] 伍星[1] 

机构地区：[1]上海交通大学振动,冲击,噪声国家重点实验室,上海200030

出　　处：《计算机应用》2004年第1期4-5,7,共3页journal of Computer Applications

基　　金：国家"十五"计划资助项目 (2 0 0 1BA2 0 4B0 5 -KHKZ0 0 0 9)

摘　　要：对由连续条件属性值和离散决策属性值组成的决策表 ,提出了一种归纳学习方法。把决策表中的连续条件属性值看作一矩阵 ,进行矩阵的奇异值分解 ,以确定决策表条件属性的数目。用模糊C均值聚类的方法对连续条件属性值进行不同聚类数目的聚类 ,得到不同聚类数目下的离散决策表 ,对这些决策表进行条件属性简化 ,从而得到不同的条件属性数目。比较矩阵奇异值分解后决策表条件属性的数目和上述不同聚类数目下的离散决策表简化后的条件属性的数目 ,并考虑决策属性的数目 ,确定最终的聚类数目。在此基础上 ,给出了由连续条件属性值和离散决策属性值组成的决策表的归纳学习方法 ,并验证了其有效性。An inductive learning approach of decision table (DT) with continuous condition attribute values (CCAVs) and discrete decision attribute values is proposed in this paper. Take all the CCAVs of DT as a matrix to do singular value decomposition (SVD) and the reduced number of condition attributes can be achieve. The approach of Fuzzy C-Means Clustering (FCM) is employed to discretize CCAVs of DT,where for distinct condition attribute,same clustering number is operated within the DT. So,different discrete decision tables are acquired with various clustering number for different operation. Reduce all of the discrete DTs and the reduced condition attribute numbers can be obtained. Then,the final clustering number of DT with CCAVs is selected by comparing the number of condition attributes calculated by SVD of matrix with those above reduced condition attribute numbers,considering the number of decision attribute of DT. On the basis of it,an inductive learning approach is presented. Finally,an example is illustrated and proves that the approach is effective.

关 键 词：奇异值分解 粗糙集 决策表 模糊C均值聚类 归纳学习 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]