基于关联度和代价敏感学习的决策树生成法  

Decision Tree Based on Correlation Degree and Cost-sensitive Learning

作　　者：孟光胜[1] 

机构地区：[1]河南科技大学林业职业学院,洛阳471002

出　　处：《科学技术与工程》2013年第5期1196-1199,共4页Science Technology and Engineering

基　　金：河南省教育厅科技攻关项目(2010B520033)资助

摘　　要：针对ID3决策树生成法中存在的缺点,通过引用属性关联度和代价敏感学习,提出了一种基于属性关联度和代价敏感学习的决策树生成法。该方法利用粗糙集理论对条件属性进行约减,在构建决策树过程中,把属性的关联程度和性价比作为选择分裂结点的依据,利用改进的信息增益方法构建代价敏感决策树。试验结果表明,该方法在分类准确度和生结点总数量上比常用决策树生成方法优越。Aiming at the shortcomings of the ID3 algorithms in decision tree generation, by introducing the attribute correlation degree and cost-sensitive learning,this paper presents the method for generationg cost-sensitive decision tree based on attribute correlation degree.The way reduce the attribute by rough set theory, and select the splitting nodes by the attribute correlation degree and performance price ratio in generationg decision tree. The improved information gain is used for Generationg Cost-Sensitive decision tree. The experimental results show that the method is superior in classification accuracy and node number than the usual algorithms in decision tree generation.

关 键 词：关联度 决策树 代价敏感学习 属性约简 

分 类 号：TP301.1[自动化与计算机技术—计算机系统结构]