Abstract | ||
---|---|---|
A novel classification algorithm named LapWCS-TSVM is proposed.To effectively exploit the geometric information from unlabeled instances via the manifold regularization term.To capture data structure information by incorporating minimum within-class scatter as in the WCS-SVM.To avoid the disadvantage of loss of information contained in the majority of training instances by adopting total margin algorithm to substitute the traditional soft margin algorithm.Validity is investigated by comparing it with related classifiers on artificial datasets, UCI datasets and face recognition datasets. Insufficient volume of supervised information is a major challenge for supervised learning. An effective method to handle this problem is semi-supervised learning, which can make full use of the geometric information embedded in unlabeled instances. In this paper, we present a novel laplacian total margin support vector machine based on within-class scatter (LapWCS-TSVM) method to deal with the semi-supervised binary classification problem. The proposed LapWCS-TSVM incorporates the total margin algorithm and the manifold regularization into WCS-SVM to help improve its performance. With the help of kernel trick, the proposed LapWCS-TSVM can be easily generalized to non-linear separable case and solved by the optimization programming of the traditional support vector machine. Experiments conducted on artificial datasets, UCI datasets and face recognition datasets show the validity of the newly proposed algorithm. |
Year | DOI | Venue |
---|---|---|
2017 | 10.1016/j.knosys.2016.12.009 | Knowl.-Based Syst. |
Keywords | Field | DocType |
Support vector machine,Total margin,Within-class scatter,Manifold regularization,Semi-supervised learning | Data mining,Margin (machine learning),Semi-supervised learning,Binary classification,Computer science,Artificial intelligence,Data structure,Facial recognition system,Pattern recognition,Support vector machine,Supervised learning,Kernel method,Machine learning | Journal |
Volume | Issue | ISSN |
119 | C | 0950-7051 |
Citations | PageRank | References |
3 | 0.36 | 18 |
Authors | ||
4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Huimin Pei | 1 | 10 | 1.82 |
Yanyan Chen | 2 | 16 | 3.59 |
Yankun Wu | 3 | 3 | 0.36 |
Ping Zhong | 4 | 40 | 11.34 |