A Study on Classifier by Weak SVM Boosting

Abstract

In this paper, an ensemble classifier constructed by AdaBoost boosting of weak SVM with Gaussian RBF kernel is studied. Each SVM classifier is trained by small part of given training data and boosted by AdaBoost technique. By using on small data set for training, it is able to reduce computational burden. The classification accuracy is improved by boosting based ensemble. We propose such weak SVM ensemble method using AdaBoost and show availability of the proposed method by some numerical simulation results. Though setting of the kernel parameter is in general done by trial and error approaches or based on some kind of prior knowledge, it is also shown that the proposed method is able to give an ensemble SVM classifier that has low sensitivity to the kernel parameter setting.