Improvement classification performance by the support vector machine ensemble

Ensemble Methods are learning algorithms that construct a collection of individual classifiers which are independent and yet accurate, and then classify a new data point by taking vote of their predictions. The support Vector Machine (SVM) presents excellent performance in solving the problems with a small number of simple, nonlinear and local minima. The combination of the Support Vector Machine with Ensemble methods has been done by Hyun-Chul Kim based on the bagging algorithm, yet it does not show high robustness for its randomicity. In this paper, by a deep investigation into the principle of the SVM and the Ensemble Method, we propose two possible ways, cross validated committees and manipulating of the input feature strategies, to construct the SVM ensemble, which provides strong robustness according to experimental results.