Center-Based Iteration Algorithm Of Pre-Extracting Support Vectors

On the basis of research on the mechanism of the support vector machine (SVM) and the distribution characteristics of support vectors, a class center-based algorithm of pre-extracting support vectors was proposed. First, a hyperplane band was constructed via the class centers of two classes of samples and the width of the band was reduced via iteration; then, the samples within the band were employed in training as an alternative to the entire training samples, thus reducing the training samples and meanwhile excluding some outliers; and finally, the algorithm proposed was tested via artificial data and UCI data, and the results of the simulation experiments indicate that when the influence on the classification precision is modest, the proposed algorithm can significantly increase the training speed.