Robust AdaBoost with SVM-Based Component Classifiers

The performance of AdaBoostSVM is easily disturbed by the outliers,and the generalization performance of the algorithm will be influenced.Outliers are the data that do not inflect the general regularity.When many misclassified outliers in the data set,the weights of the outliers usually keep increasing during the training procedue of AdaBoostSVM.Thus the generalization performance of the generated classifier gets worse.Concerning this problem,RAdaBoostSVM is proposed to improve the performance of AdaBoostSVM by approximating the misclassified training data over-weighted by the center of its several neighboring points.Compared to AdaBoostSVM,RAdaBoostSVM is more robust to the outliers and suitable to classify the noisy data.Experimental results on several benchmark data sets demonstrate the efficiency of the proposed algorithm.