基于权衡因子的决策树优化算法  

Optimized algorithm of decision tree based on weighting factor

作　　者：董跃华[1] 刘力[1] 

机构地区：[1]江西理工大学信息工程学院,江西赣州341000

出　　处：《江西理工大学学报》2015年第5期90-97,共8页Journal of Jiangxi University of Science and Technology

基　　金：国家自然科学基金资助项目(71061008)

摘　　要：通过分析ID3算法的多值偏向问题和传统ID3改进算法中出现的主观性等问题,提出了一种基于权衡因子的决策树优化算法.该优化算法通过引入能够反映属性之间相互依赖关系的权衡因子,对取值个数最多的属性的划分权重重新进行权衡,以完成对ID3算法的改进.实例验证和标准数据集UCI上的实验结果表明,当数据集中属性的取值个数不相同时,优化后的ID3算法能够解决多值偏向问题,在构建决策树的过程中,优化后的ID3算法既能提高平均分类准确率,又能减少平均叶子节点数.Through the analysis of the issues of multivalue bias in the ID3 algorithm and subjectivity of the optimized traditional ID3 algorithm, an improved algorithm of decision tree based on weighting factor is put forward. The new algorithm introduces the weight factor that reflects the mutual relationship between the attributes. The ID3 algorithm is improved by redistricting the weight of attributes which has most values. The experiments on UCI data sets show that the optimization ID3 algorithm can overcome muhivalue bias when the values of different attributes in data set are not the same. This algorithm not only improves the accuracy of average classification, but also reduces the number of average leaf nodes in the process of constructing a decision tree.

关 键 词：ID3算法 多值偏向 权衡因子 决策树 权重 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]