摘要: |
针对SVM法线特征筛选算法仅考虑法线对特征筛选的贡献,而忽略了特征分布对特征筛选的贡献的不足,在对SVM法线算法进行分析的基础上,基于特征在正、负例中出现概率的不同提出了加权SVM法线算法,该算法考虑到了法线和特征的分布。通过试验可以看出,在使用较小的特征空间时,与SVM法线算法和信息增益算法相比,加权SVM法线算法具有更好的特征筛选性能。 |
关键词: 特征筛选 支持向量机 加权SVM法线算法 文本分类 |
DOI: |
分类号: |
基金项目:西北农林科技大学人才基金项目(01140401;01140402) |
|
An improved feature selection algorithm for linear SVM classifiers |
|
Abstract: |
SVM normal feature selection algorithm only considers the contribution made to feature selection by the normal,ignoring the contribution made by the distribution of features totally.In this paper,based on the analysis of SVM normal algorithm,and the difference between the probability that attributes occurs in positive and negative samples,a weighed SVM normal algorithm,which considers both the distribution and the normal was presented.The experiment results show that,compared with SVM normal algorithm and information gain algorithm,when a small feature space is applied,the weighed SVM normal algorithm has better feature selection performance than SVM normal algorithm and information gain algorithm. |
Key words: feature selection SVM weighed SVM normal algorithm text categorization |