Pattern recognition with novel support vector machine learning method
Mikko I. Lehtokangas · European Signal Processing Conference · 2000
The concept of optimal hyperplane has been recently proposed in the context of statistical learning theory. The important property of an optimal hyperplane is that it provides maximum margins to each class to be separated. Obviously, such a decision boundary is expected to yield good generalization. Currently, the support vector machines (SVM) are probably one of the very few models (if not the only ones) that make use of the optimal hyperplane concept. In this study we investigate the basic SVM method and point out some problems that may arise especially in large scale problems with abundant data. Moreover, we propose a novel SVM type method that aims to avoid the problems found in the basic method. The experimental results demonstrate that the proposed method can give very good classification performance. However, the results also point out another potential problem in the SVM scheme which should be considered in the future studies.