关键词: 支持向量机, 大规模分类, 边界样本, 聚类

Abstract: Support Vector Machine（SVM）is a powerful instrument for solving pattern classification problem, but it is not suitable for large-scale data, due to the drawbacks of slow training speed, large computational cost and low generalization. An accurate support vector machine algorithm is proposed, which uses training samples lying close to the separation boundary. First of all, K-means clustering is performed to the initial training data, and then the boundary samples are selected in each cluster by K-nearest neighbor algorithm, two cluster factors, the degree of mixing and support, are defined to determine the boundary width. These boundary samples are then used in the training of the SVM classifier. The experiments on some benchmark datasets show that the proposed method not only makes computational complexities decreased, but also makes classification power of traditional SVM invariant.

Key words: Support Vector Machine（SVM）, large-scale classification, boundary samples, clustering