A research and improvement on support vectors pre-extracting method based on vector projection

Abstract

For the support vectors pre-extracting method based on vector projection, the extraction procedure of the bound vector set has a higher computation cost and needs more bound vectors to be selected to include the support vectors. For this problem, we propose a novel bound vectors extracting method by studying the distribution characteristics of vector projection of samples. This method only relies on the projection values of samples and avoids complex computation. Furthermore, a binning method is adopted to determine the borders between different type projection regions more exactly. In this way we can extract fewer bound vectors for training. Experiments show that the proposed method is almost as accurate as standard SVM but is much faster and effective.