一种基于规范互信息和动态冗余信号识别技术的特征选择方法  

A normal mutual information and dynamic redundant signal identification technique based feature selection method

作　　者：陈圣[1] 熊钦[2] 

机构地区：[1]华中科技大学电子与信息工程系,湖北武汉430074 [2]华中科技大学电子科学与技术系,湖北武汉430074

出　　处：《电子设计工程》2012年第18期142-144,147,共4页Electronic Design Engineering

摘　　要：为了实现对模式识别、信号处理等领域中数据的有效表达,提出了一种基于规范互信息和动态冗余信号识别技术的特征选择方法。该方法采用规范互信息对特征相关性和冗余性进行测量,并通过一种动态冗余信号识别技术在特征全集中进行冗余特征的筛选。分类实验结果表明所提特征选择方法性能优于典型的特征选择方法。In order to effectively express the data in the field of pattern recognition and signal process, a normal mutual information and dynamic redundant signal identification technique based feature selection method is proposed. This method applies normal mutual information to measure relevance and redundancy among features and the class, and employs an dynamic redundant signal identification technique to eliminate redundant features in the whole feature set. Experimental results on classification show that the proposed method outperforms those representative feature selection methods.

关 键 词：特征选择 规范互信息 冗余信号识别 分类 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]