Multicategory Nonparallel Proximal Support Vector Machine
Xubing Yang, Songcan Chen, Zhisong Pan · 2007
We propose a multicategory classifier, that is, Multicategory Nonparallel Proximal Support Vector Machine (MNPSVM), which is in the spirit of Proximal SVMs via Generalized Eigenvalues (GEPSVM).Difference from GEPSVM lie in: 1) MNPSVM keeps the genuine rather than approximate geometrical interpretation of the nonparallel proximal SVMs; 2) each nonparallel plane of MNPSVM is generated by its corresponding standard eigenvalue problems, instead of nowadays generalized eigenvalue problems.The effectiveness is demonstrated by tests on synthetic and real data sets.Furthermore, we also discuss its efficiency in experiment section and conclude that MNPSVM is far higher than that of both GEPSVM and SVM.