Logistic Regression with Robust Bootstrapping

The bootstrap method is a widely used technique for statistical learning and inference. However, its performance can be dramatically affected by outliers in the training data. In this paper, we propose two robust bootstrapping algorithms for logistic regression. These methods have several compelling advantages: they alleviate the dependency on fine-tuning the hyperparameters, and can still remain robust against highly contaminated data. The performance of the proposed algorithms is evaluated on synthetic and real-world data sets.