Abstract: Incremental support vectors determine the generalization performance of incremental support vector machine,so their choosing is significant We proposed a novel strategy to choose the incremental support vectors by computing their probabilities. The method constructs an appropriate separating hyperplane for each training sample, which goes through the corresponding training sample point. And then the probability of a sample to be incremental support vector is estimated by the separating rate of the corresponding hyperplane. Those samples with higher probabilities are chosen as the incremental support vectors to train and update the incremental support vector machine. Comparative numerical experiments of our method against the existing methods show that our method has outstanding advantage in the training efficiency without deteriorating the generalization performance of incremental support vector machine.

Key words: Incremental support vector machine, Incremental support vector, Support separating hyperplane, Support separating rate