Extraction of the Reduced Training Set Based on Rough Set in SVMs.

In SVMs, the data points lying in the interactive regions of two classes are very important to form the hyperplane and difficult to be classified. How to select the reduced training set only including the interactive data points is one of the important issues. There are many methods by which the easy misclassified training data are selected to speed up training. The extraction method of the reduced training set is proposed by using the boundary of rough set. Firstly, for two-class problem, the entire training set is partitioned into three regions: the region only containing the positive samples, the region only composed of the negative samples and the boundary region including not only the positive samples but also the negative ones. Secondly, the boundary region is the intersection of two classes and selected to train SVMs. Thirdly, the two-class and multi-class problems are used to verify the feasibility of the proposed SVMs. The experimental results on the classic benchmark data set of machine learning show that the proposed learning machines can downsize the number of training data and hardly influence on their generalization abilities. [ABSTRACT FROM AUTHOR]