A effective classified algorithm of support vector machine with multi-representative points based on nearest neighbor principle

In this paper, a classification algorithm of the support vector machine (SVM) with multi-representative points is studied, which aims at reducing the long training time for large scale data and improving classification accuracy for a complicated problem. With regarding traditional SVM as a 1 nearest neighbor (1NN) classifier in which only one representative point is selected for each class, the idea is to divide the training set into c subsets, then several SVMs are trained by them and every training result can be chosen as one representative point. A dividing method is given where the positive and negative examples are divided into several clusters respectively, which are combined into subset pairs according to calculating the distance between the positive and the negative clustering centers. Finally the classified algorithm was designed as the nearest neighbor algorithm in which c representative points are chosen for each class. The numerical experiments show that our algorithm not only can reduce the training time notability but also improve the classification accuracy to a certain extent