[1]Li Mengmeng,Shang Zhigang,Li Zhihui.Fast Method to Filter Support Vectors Combined with Operation of Projection and Nearest Neighbors’ Selection[J].Journal of Zhengzhou University (Engineering Science),2017,38(03):49-53.[doi:10.13705/j.issn.1671-6833.2016.06.003]
Copy
Journal of Zhengzhou University (Engineering Science)[ISSN
1671-6833/CN
41-1339/T] Volume:
38卷
Number of periods:
2017 03
Page number:
49-53
Column:
Public date:
2017-05-28
- Title:
-
Fast Method to Filter Support Vectors Combined with Operation of Projection and Nearest Neighbors’ Selection
- Author(s):
-
Li Mengmeng; Shang Zhigang; Li Zhihui
-
School of Electrical Engineering, Zhengzhou University, Zhengzhou, Henan 450001
-
- Keywords:
-
- CLC:
-
-
- DOI:
-
10.13705/j.issn.1671-6833.2016.06.003
- Abstract:
-
To reduce computational burden and improve operation efficiency of support vector machine (SVM) while ensuring classification accuracy,a fast method to filter support vectors combined with operation of projection and nearest neighbors’ selection was proposed.Considering the global characteristics of Fisher projection,it could be viewed as the approximate normal directions of SVM optimal hyperplane and filtered out a large number of non-support-vectors in this direction.The samples near the classification obtained boundary were regarded as alternative support vectors set.Neighborhood operation was combined to solve the problem that some support vectors might be filtered out mistakenly regardless of the local structure information.A number of nearest neighbors of the alternative support vectors were selected backward from the samples space to update and expand the alternative support vectors set.The sets was treated as the SVM input.The experimental results on several UCI standard data sets showed that the fast method had good generalization performance and reduced the computational burden effectively under the premise of fully guaranteed classification accuracy.