A Fast SVM Training Algorithm Based on a Decision Tree Data Filter

Abstract

In this paper we present a new algorithm to speed up the training time of Support Vector Machines (SVM). SVM has some important properties like solid mathematical background and a better generalization capability than other machines like for example neural networks. On the other hand, the major drawback of SVM occurs in its training phase, which is computationally expensive and highly dependent on the size of input data set. The proposed algorithm uses a data filter to reduce the input data set to train a SVM. The data filter is based on an induction tree which effectively reduces the training data set for SVM, producing a very fast and high accuracy algorithm. According to the results, the algorithm produces results in a faster way than existing SVM implementations (SMO, LIBSVM and Simple-SVM) with similar accurateness.