Selective SVM Ensemble Based on Dynamic Rough Set
Tao Chen · Jisuanji fangzhen · 2012
Ensemble is an effective method to improve generalization performance of SVM.Individual SVM's accuracy and the difference between SVMs are two key factors to affect the generalization performances.Selective SVM ensemble based on dynamic rough set was presented to improve the generalization ability of SVM.First,the training samples were disturbed by using conventional Boosting algorithm.A dynamic reduction technology,which integrates genetic algorithm and resample method,was used to acquire the reducted sets that have stable and good generalization ability.Best individual was selected according to generalization error of SVM based on the validate set based on KFCM.Finally,the selected members were ensembled nonlinearly by SVM.The experiments show that the algorithm has higher generalization performance and lower time and space complexity.It is a higher effect ensemble algorithm.