FAST TRAINING OF LARGE DATA SET ON SOM-SVM FOR PATTERN RECOGNITION

摘要

Support Vector Machine (SVM) has been widely used in classification or regression problems and showed the high advantage compared to the other learning machines. However, despite the prominent properties of SVM, it is known that the computational cost is highly dependent on the data size because it is formulated as a quadratic programming (QP) problem. In this paper, we propose a new method, integrating a Self-Organizing Maps (SOM) with a Support Vector Machine (SVM), which is specifically designed for handling large data set. SOM distributes the large data set to the units including small data sets and these units are used to develop the classifiers on SVM. Experiments on a couple of well known data sets verify the computation time can be significantly reduced without any critical decrease in the classification accuracy.