Protein secondary structure prediction based on multi-SVM ensemble

To improve the performance of secondary structure prediction, a multi-SVM ensemble was applied, bagging was used to resample the training dataset. The SVM ensemble was made of two-layer, one is composed by three SVM network decided by winner-take-all, the other is a ensemble network composed of five classifier decided by majority voting. Seven-fold cross-validation test on RS126 dataset indicated that the multi-SVM ensemble could achieve better performance on secondary structure prediction.