基于BSO-OS算法的两阶高维数据特征选择  

Two-stage feature selection for high-dimensional data based on BSO-OS algorithm

作　　者：田浩楠 周晖[1] TIAN Hao-nan;ZHOU Hui(School of Information Science and Technology,Nantong University,Nantong 226019,China)

机构地区：[1]南通大学信息科学技术学院,江苏南通226019

出　　处：《计算机工程与设计》2020年第3期695-700,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(61501264)。

摘　　要：针对高维数据特征选择问题,提出基于BSO-OS的两阶特征选择算法。采用BSO-OS算法搜索特征子集,产生分类性能较好、特征数量较少的特征子集;对上述所选特征子集中分类精度最高的特征子集采用FAMIR算法,去除不相关和冗余特征,产生分类性能更好、特征数量更少的特征子集。对6个高维数据集进行实验,实验结果表明,所提算法选择的特征子集相较一阶方法和传统的两阶方法具有更好的分类性能。To address the problem of feature selection for high-dimensional data,a two-stage feature selection algorithm based on BSO-OS(brain storm optimization in objective space)was proposed.BSO-OS algorithm was used to search for feature subsets with better classification accuracy and fewer features.FAMIR(fast approximate joint mutual information)was applied to the feature subset with the highest classification accuracy found in the first stage to eliminate irrelevant and redundant features,so that feature subsets with much better classification accuracy and much fewer features were obtained.Experimental results on the six high-dimensional datasets demonstrate that the feature subset found using the proposed two-stage method is better than the one using other algorithms in terms of classification accuracy.

关 键 词：特征选择 高维数据 BSO-OS算法 FAMIR算法 两阶算法 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]