An improved reduced support vector machine

The reduced support vector machine (RSVM) was proposed to overcome the computational difficulties as well as to reduce the model complexity in generating a nonlinear separating surface for a massive data set. However, it selects `support vectors´ randomly from the training set, this will effect the result. To overcome this shortcoming of RSVM, an improved RSVM algorithm is presented in this paper. First of all, we calculate the average of relative distance for each sample point in every class; and then use percentile to deal with unbalanced sample and remove the outliers form margin vectors, so the representative vectors as `support vectors´ were extracted; finally, we apply the RSVM on these representative vectors. Because we reduce the effect of unbalanced sample and outliers, and apply the representative vectors as `support vectors´, so the new algorithm improves the ability of RSVM to classify and the training speed of C-SVM .