Clustering-Based Geometric Support Vector Machines

Training a support vector machines on a data set of huge size may suffer from the problem of slow training. In this paper, a clustering-based geometric support vector machines (CBGSVM) was proposed to resolve this problem, initial classes are got by k-means cluster, then develop a fast iterative algorithm for identifying the support vector machine of the centers of all subclasses. To speed up convergence, we initialize our algorithm with the nearest pair of the center from opposite classes, and then use an optimization-based approach to increment or prune the candidate support vector set. The algorithm makes repeated passes over the centers to satisfy the KKT constraints. The method speeds up the training process fast comparing with standard support vector machines under the almost same classification precision.