Support vectors pre-extracting method based on boundary distance

摘要

A novel method called boundary distance is proposed for pre-extracting support vectors. It first calculates the distance between the sample and the other class sample. According to sort distance, the less distance samples, and nearest neighbor samples in the other class, are used as boundary samples. As the boundary samples include most support vectors, it greatly declines the training time without any loss of classification accuracy. Experiments on two artificial data sets and UCI standard data set show that the proposed method is effective.