Reduce the number of support vectors by using clustering techniques
Quang-Anh Tran, Qianli Zhang, Xing Li · 2004
A serious problem of support vector machine (SVM) is its low classifying speed. The speed depends on the number of support vectors. The clustering SVM, proposed in this paper, is a new method to reduce the number of support vectors. The method uses a k-means clustering technique to assign the data of each class to k groups, then we train the SVM based on a new dataset consist of the central vectors of each group. The value of k is the upper bound of the number of support vector in the class. Experiment results demonstrate that our method can control the tradeoff between the classifying speed and the performance of SVM.