A support vector machine training algorithm based on cascade structure

To apply Support Vector Machine (SVM) to deal with larger training data, a training algorithm based on cascade structure is proposed, which is not based on solving a complex quadratic optimization problem but divide and conquer strategy. Cascade structure is applied to reduce the number of training data in each training process, and multiple SVM classifiers are obtained which represented learning results of every training subset. The support vector sets obtained correspondingly are combined and added back into training subsets as feedbacks. Feedbacks are necessary when considering the problem that the learning results are subject to the distribution state Of the training data in different subsets. The experimental results on UCI dataset show that the proposed training algorithm is able to deal with larger scale learning problems, and the suitable feedback strategy makes the learning accuracy more satisfying and less computation time cost compared with standard cascade SVM algorithm.