A Comparison of Support Vector Machines Ensemble for Classification

Support Vector Machines (SVM) is characteristic of processing complex data and high accuracy. An ensemble of classifiers often results in better performance than any single classifier in the ensemble. In this paper, bagging, boosting, multiple SVM decision model (MSDM) and heterogeneous SVM decision model (HSDM) of SVM ensemble are compared on four data sets. For boosting, a novel strategy for weight updating, doubling the misclassified samples, is introduced to AdaBoostMl, which we call dboosting. Experiment results show that dboosting with SVM outperforms other methods in term of accuracy. HSDM can also improve the accuracy too. Bagging is not obvious and MSDM performs worst.