A Novel Proximal Support Vector Machine and Its Application in Radar Target Recognition
Tao Xiaoyan, Xia Jingbo, Rui Zhang · 2006
The samples are assumed to distribute normally in the solution of the standard proximal support vector machine (PSVM). But in many application problems, the data set for each class is generally unbalanced, where a poor performance can be gotten by PSVM. For this, a novel PSVM is presented, namely the modified PSVM (MPSVM). By adding a new diagonal matrix in the primal optimization problem, the new algorithm assigns the different penalty coefficients to the positive and negative samples respectively. Therefore the samples in different classes can make different contributions to the learning of the decision surface. Based on the sufficient experimental results on the UCI datasets, MPSVM is also applied to the measured radar range profile images and the results illustrate the effectiveness of the proposed method.