信息论特征选择算法的改进

特征选择在模式识别技术中起着非常重要的作用，已有多种特征选择的方法，但用信息论的方法进行特征选择还是一个新的课题，MIFS算法和MIFS-U算法都是近似算法，随着输入特征的增加，特征选择性能逐渐下降．本文通过研究这两种算法，提出一种改进方法，在运算量几乎不增加的情况下，提高这两种算法的特征选择性能。

An Improved Method for Feature Selection Algorithm Based on Information Theory

Feature selection plays an important role in classification problems such as pattern recognition, there are a lot of methods to select feature, but the methods based on information theory to select feature is a new subject yet.MIFS methods and MIFS-U methods are all approximate methods, their feature selection performance will be decreased with the increasing of selected feature number. This paper describes a method to modify the two kinds of methods to improve their feature selection performance without more calculation increasing.