摘要
通过寻找一个最优的特征子集,特征选择可以降低计算复杂度,提高分类精度以及结果的可理解性。提出基于大间隔信息粒化的特征选择算法,通过聚类等方式对原始数据进行单类信息粒化,然后在粒化的基础上构造了模糊间隔和类间隔2个评价指标进行特征评价。并分别在不同的数据上验证了这种特征选择方法的有效性,实验结果表明,基于大间隔粒计算的特征选择算法效果要优于其他的大间隔特征算法。
Feature selection is used to find an optimal subset to reduce computational cost, increase classification accuracy and improve result comprehensibility. In this paper, we introduced a feature selection technique based on information gran- ularity and large margin. Firstly, we operated the information granularity on raw data, and then based on information granu- larity we proposed fuzzy margin and class margin as the feature evaluation functions. The effectiveness of the proposed method was validated by experiments on different data sets. Experimental results show that the proposed technique has bet- ter performance than the other margin based feature selection methods.
出处
《重庆邮电大学学报（自然科学版）》
北大核心
2010年第5期641-647,共7页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
基金
国家自然科学基金(10978011/A030402)~~
关键词
特征选择
信息粒化
大间隔
模糊C均值
非负矩阵分解
feature selection
information granularity
large margin
fuzzy C-means （ FCM ）
non-negative matrix factorization（NMF）
