[show abstract][hide abstract] ABSTRACT: Support vector machine(SVM) has been widely used for its outstanding performance, but, it still has flaws. One of them is that SVM is unit sensitive. In this paper, we analyze how will the different units effect the SVM. Then, we propose a preprocess method not only to conquer this flaw, but also improve the generalization precision of SVM. The preprocess method is base on decision tree(DT). The idea is using DT to train the data first, then, scaling the data base on the outcome decision tree. Finally, SVM is adapted on the new data for training and prediction. Experimental results on real data show remarkable improvement of generalization precision.
Computer Science and Service System (CSSS), 2011 International Conference on; 07/2011
[show abstract][hide abstract] ABSTRACT: A novel support vector machine (SVM) algorithm for regression problems is proposed in this paper. Each pattern in the original training set is converted into a pair of patterns, which are labeled by 1 and -1, respectively. Therefore, the regression problem can be considered as a classification problem. By optimizing the obtained decision function, the model output of unknown samples can be estimated. Experimental results show the proposed method works well, and in many cases it produces less support vectors than the normal support vector regression (SVR) machine.
Computer Engineering and Technology (ICCET), 2010 2nd International Conference on; 05/2010