连续值属性决策表中的知识获取方法

Knowledge Acquisition from Decision Tables Containing Continuous-Valued Attributes

提出了一种从连续值属性决策表中获取知识的方法KACVA(Knowledge Acquisition from decision tablescontaining Continuous-Valued Attributes).该方法将经典粗糙集理论对数据空间的等价划分转换为相似划分,把传统粗糙集理论中正域的表示方法扩充到连续值属性决策表中;通过计算连续值属性决策表中各条件聚类对决策类的分类能力,生成决策规则.不同数据集的实验测试结果表明:对连续值属性决策表中的知识获取,KACVA方法与传统的粗糙集相关知识获取方法及C4.5决策树分类方法相比,有更高的分类准确率.

An approach for Knowledge Acquisition from decision tables containing Continuous-Valued Attributes （KACVA） is developed. The equivalence partition in the classical rough sets theory is converted into a similarity partition. A novel representation method of the positive region in decision tables with continuous-valued attributes is built. Through calculating classification abilities of each conditional cluster to decision classes, decision rules in decision tables containing continuous-valued attributes are generated. Experimental evaluation on different data sets shows that the KACVA algorithm has the better performance in the classification accuracy comparing with the knowledge acquisition approaches under classical rough sets theory and the decision tree approach, C4-5, in processing decision tables with continuous-valued attributes.