A preprocessing method of AdaBoost for mislabeled data classification

AdaBoost is one of the most popular algorithm for classification and has been successfully used for text classification, face detection and tracking. However noise sensitivity is regarded as a major disadvantage and previous works show that AdaBoost will be overfitting when dealing with the data sets with noisy data. To improve the noise tolerance of conventional AdaBoost, this paper proposed a preprocessing method of AdaBoost for mislabeled data to find the noisy data and correct it. Further decision stump is selected as the weak learner of the AdaBoost algorithm for classification. The comparison of simulation results between conventional AdaBoost and the method proposed in this paper shows that the proposed algorithm has improved testing accuracy of the data sets with the noisy data.