基于类别比例因子和类内均分度的χ^2统计改进  

Improvement χ^2 of Statistics Based on the Category Scale Factor and Average Distribution Inner Category

作　　者：张瑜[1] 张德贤[1] 

机构地区：[1]河南工业大学信息科学与工程学院,河南郑州450001

出　　处：《电子科技》2010年第12期70-72,共3页Electronic Science and Technology

基　　金：河南省创新性科技团队建设基金资助项目(094200510009)

摘　　要：针对2χ统计特征选择方法的两大局限:对低文档频的特征选择不合理,以及过分强调那些在指定类低频出现,而在其他类中高频出现的特征项在该类中的权重。提出基于类别比例因子与类内均分度的2χ统计特征选择的改进方法。实验结果表明,改进方法的分类效果优于传统方法。The χ^2statistical method has two defects.One has reduced the weight of the low-frequency terms and the other has increased the weight of the characteristics in the designated class.This paper proposes an improved χ^2statistical approach based on the category scale factor and average distribution inner category.A contrastive experiment is carried out and the results show that improved χ^2statistics is superior to traditional statistical in feature selection.

关 键 词：特征选择 2χ统计 类别比例因子 类内均分度 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]