What makes multi-class imbalanced problems difficult? An experimental study

作者：

Highlights：

• Difficulty factors in multi-class imbalanced data are experimentally studied.

• A strong influence of class overlapping related to types of class inter-relation.

• High impact of the class size configuration (multi-majority — most difficult).

• The gradual configurations with classes of intermediate sizes play special roles.

摘要

•Difficulty factors in multi-class imbalanced data are experimentally studied.•A strong influence of class overlapping related to types of class inter-relation.•High impact of the class size configuration (multi-majority — most difficult).•The gradual configurations with classes of intermediate sizes play special roles.

论文关键词：Imbalanced data,Classification,Learning from multiple classes,Data difficulty factors

论文评审过程：Received 26 January 2021, Revised 16 January 2022, Accepted 19 March 2022, Available online 2 April 2022, Version of Record 11 April 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.116962