Resampling methods versus cost functions for training an MLP in the class imbalance context

Abstract

The class imbalance problem has been studied from different approaches, some of the most popular are based on resizing the data set or internally basing the discrimination-based process. Both methods try to compensate the class imbalance distribution, however, it is necessary to consider the effect that each method produces in the training process of the Multilayer Perceptron (MLP). The experimental results shows the negative and positive effects that each of these approaches has on the MLP behavior