Incremental support vector machine learning in the primal and applications

Most algorithms of support vector machines (SVMs) operate in a batch mode. However, when the samples arrive sequentially, batch implementations of SVMs are computationally demanding due to the fact that they must be retrained from scratch. This paper proposes an incremental SVM algorithm that is suitable for the problems of sequentially arriving samples. Unlike previous SVM techniques, this new incremental SVM learning is implemented in the primal and it shows that the primal problem can be efficiently solved. The effectiveness of the proposed method is illustrated with several data sets including faces, handwritten characters and UCI data sets. These experiments also show that the proposed method is competitive with previously published methods. In addition, the application of the proposed algorithm to leave-one-out cross-validation is demonstrated.