Abstract
Ensemble methods have been known to improve the prediction accuracy over the base learning algorithms. AdaBoost is well-recognized for this in its class. However, it is susceptible to overfitting the training instances corrupted by class label noise. This paper proposes a modification of AdaBoost that is more tolerant to class label noise, which further enhances its ability to boost the prediction accuracy. Particularly, we observe that in Adaboost, the weight-hike of noisy examples can be constrained by careful application of a cut-off in their weights. We study the characteristics of our technique empirically using some artificially generated data set. We also corroborate this on a number of data sets from UCI repository [1]. In both experimental settings, the results obtained affirm the efficiency of our approach. Finally, some of the significant characteristics of our technique related to noisy environments have been investigated.