Deep learning regularization in imbalanced data

Abstract

Deep neural networks are known to have a large number of parameters which can lead to overfitting. As a result various regularization methods designed to mitigate the model overfitting have become an indispensable part of many neural network architectures. However, it remains unclear which regularization methods are the most effective. In this paper, we examine the impact of regularization on neural network performance in the context of imbalanced data. We consider three main regularization approaches: L{1}, L{2}, and dropout regularization. Numerical experiments reveal that the L{1} regularization method can be an effective tool to prevent overfitting in neural network models for imbalanced data. Index Terms-regularization, neural networks, imbalanced data. © 2020 IEEE.