Reduced Support Vector Machine Based on Margin Vectors

Reduced Support Vector Machine (RSVM) was proposed as an alternate of the standard SVM. Motivated by resolving the difficulty on handling large data sets using SVM, it pre-extracts a subset of data as `support vectors´ and solves a smaller optimization problem. But it selects `support vectors´ randomly from the training set, and this will affect the result. A new method called reduced support vector machine based on margin vectors is presented in this paper, some margin vectors were extracted as `support vectors´ via center distance ratio, then were applied in the RSVM . The new method can be used to unbalanced data and reduce the effects of outliers. So the new method improves the ability of RSVM to classify and the training speed of SVM greatly.