Improving the Robustness of Bagging with Reduced Sampling Size

Bagging is a simple and robust classification algorithm in the presence of class label noise. This algorithm builds an ensemble of classifiers by bootstrapping samples with replacement of size equal to the original training set. However, several studies have shown that this choice of sampling size is arbitrary in terms of generalization performance of the ensemble. In this study we discuss how small sampling ratios can contribute to the robustness of bagging in the presence of class label noise. An empirical analysis on two datasets is carried out using different noise rates and bootstrap sampling sizes. The results show that, for the studied datasets, sampling rates of 20% clearly improve the performance of the bagging ensembles in the presence of class label noise.