基于互信息最大化和特征聚类的特征选择

Feature Selection Based on Mutual Information Maximization and Feature Clustering

提出一种互信息最大化和特征聚类相结合的特征选择法,并将其应用于邮件识别。通过互信息最大化从原始特征空间中选择次优特征子集,借助于特征空间的聚类来剔除冗余特征,从而实现特征空间的再次降维。实验结果表明该方法是一种有效的特征选择法。

Proposes a feature selection method based on mutual information maximization and feature clustering and applies to mail recognition.Suboptimal feature subset is selected from original feature space through mutual information maximization and then redundant features are removed with the clustering feature space to achieve reduction of the number of features again. Experimental results show that the method is an effective method of feature selection.