特征选择方法的研究和改进  

Investigation and improvement of feature selection method

作　　者：张永[1] 陈思睿[1] 杨志勇[1] 贾桂霞[1] 

机构地区：[1]兰州理工大学计算机与通信学院,甘肃兰州730050

出　　处：《兰州理工大学学报》2006年第5期92-95,共4页Journal of Lanzhou University of Technology

基　　金：甘肃省自然科学基金(3ZS042-B25-014)

摘　　要：目前大部分特征选择方法都是基于对每个特征按照一定的算法进行重要性赋值,然后选取一定数目的特征作为最后的类别中心向量.针对由于初始特征过多而造成分类精度不高和分类时间过长的缺点,引进了粗糙集中的依赖度函数对特征的重要性进行标定,从而解决了特征过多造成的缺点.最后通过实验验证了该算法,证明该方法是可行的.At present, in the most part of feature selection methods an importance value is to be assigned to each feature according to a definite algorithm, and then some features are selected as the final categorization central vector. Aimed at their disadvantage of low categorization accuracy and excessive great time consumption for the categorization due to the existence of excessive initial features, the importance calibration of the features was conducted by using the dependence function in rough set, so that the problem caused by excessive features was overcome. Finally, this algorithm was verified experimentally to be feasible.

关 键 词：特征选择 属性约简 分类 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]