Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
4948570 | Neurocomputing | 2016 | 11 Pages |
Abstract
We formulate K-category non-parallel classifiers by using “One-Versus-All” (OVA) approach: Multicategory Generalized Eigenvalue Proximal Support Vector Machine (MGEPSVM) and Multicategory Improved Generalized Eigenvalue Proximal Support Vector Machine (MIGEPSVM). These classifiers generate K nonparallel decision surfaces for K classes where each surface is closest to its corresponding class and the farthest from the rest of the classes. However, in some cases, this approach leads to poor performance due to the resultant of two unbalanced classes. To minimize the effect of unbalanced classes, we proposed Weighted MGEPSVM (WMGEPSVM) and Weighted MIGEPSVM (WMIGEPSVM) where the weight factor is determined by using proposed modified balancing technique. In this paper, we also propose K-category non-parallel classifiers by using “One-Versus-One” (OVO) approach: Multicategory Generalized Eigenvalue Proximal Support Vector Machine (OVO-MGEPSVM) and Multicategory Improved Generalized Eigenvalue Proximal Support Vector Machine (OVO-MIGEPSVM). To check the robustness of the model numerical experiments have been carried out on twelve different benchmark datasets. Experimental results indicate that WMGEPSVM and WMIGEPSVM improve the testing accuracy of MGEPSVM and MIGEPSVM, while maintaining the computation time. WMGEPSVM and MGEPSVM are superior to multi-class SVM, MLSTSVM and comparable with multicategory Proximal Support Vector Machine (PSVM).
Related Topics
Physical Sciences and Engineering
Computer Science
Artificial Intelligence
Authors
Deepak Kumar, Manoj Thakur,