Novel regularization method for the class imbalance problem

作者：

Highlights：

• Class imbalance problems cause performance decline by making the model skewed.

• Regularization methods using the aspects of the output of the network are proposed.

• Proposed methods can improve performance when the training data is not balanced.

• The combination of proposed regularization methods can achieve further improvements.

摘要

•Class imbalance problems cause performance decline by making the model skewed.•Regularization methods using the aspects of the output of the network are proposed.•Proposed methods can improve performance when the training data is not balanced.•The combination of proposed regularization methods can achieve further improvements.

论文关键词：Regularization,Class imbalance,Sentence classification,Image classification

论文评审过程：Received 2 March 2020, Revised 13 July 2021, Accepted 22 September 2021, Available online 2 October 2021, Version of Record 14 October 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.115974