集合CHI与IG的特征选择方法

通过分析特征词与类别间的相关性,在原有卡方特征选择和信息增益特征选择的基础上提出了两个参数,使得选出的特征词集中分布在某一特定类,并且使特征词在这一类中出现的次数尽可能地多;最后集合CHI与IG两种算法得到一种集合特征选择方法(CCIF)。通过实验对比传统的卡方特征选择、信息增益和CCIF方法,CCIF方法使得算法的微平均查准率得到了明显的提高。

Collective CHI and IG feature selection method

In order to make the selected features distribute intensively in a certain class and make features appear in that certain class as many as possible, this paper added the two adjusted parameters to the originally traditional CHI-square feature selection and IG feature selection method through analyzing the relevance between features and classes. Then it proposed a collective feature selection methodCCIFby combining with CHI and IG feature selection. Experiments show that CCIF improves the Micro-Pmore apparently by comparing with the traditional CHI and IG feature selection method.