A novel p-margin kernel learning machine (p-MKLM) with magnetic field effect is proposed in allusion to pattern classification problem in this paper. In the Mercer induced feature space, p-MKLM can effectively resolve one-class/binary classification problems. By introducing an adjustable magnetic field density q, the basic idea of p-MKLM is to find an optimal hyperplane with magnetic field effect such that the distance between one class and the hyperplane is as small as possible due to the magnetic attractive effect, while at the same time the margin between the hyperplane and the other class is as large as possible due to magnetic repulsion, thus implementing both maximum between-class margin and minimum within-class volume so as to improve the generalization capability of the proposed method. To construct such a hyperplane with magnetic field effect, it is only needed to solve a convex quadratic programming problem which can be effciently solved with the off-the- shelf software packages for training learning machine. Experimental results obtained with benchmarking datasets show that the proposed algorithm is effective and competitive to other related methods in such cases as two-class and one-class (or novelty detection) pattern classification respectively.
Jianwen Tao, Shitong Wang, Wenjun Hu, Wenhao Ying. p-Margin Kernel Learning Machine with Magnetic Field Effect for Both Binary Classification andNovelty Detectio. International Journal of Software and Informatics, 2010,4(3):305~324Copy