EUSBoost: Enhancing ensembles for highly imbalanced data-sets by evolutionary undersampling
作者：
Highlights：
• A new ensemble solution for the class-imbalance problem is proposed.
• The accuracy of the base classifiers is enhanced by evolutionary undersampling.
• A diversity exploitation mechanism is developed, which is useful to deal with imbalance.
• EUSBoost outperforms the state-of-the-art ensembles in highly imbalanced data-sets.
• EUSBoost behavior is explained via kappa-AUC error diagrams.
摘要
Highlights•A new ensemble solution for the class-imbalance problem is proposed.•The accuracy of the base classifiers is enhanced by evolutionary undersampling.•A diversity exploitation mechanism is developed, which is useful to deal with imbalance.•EUSBoost outperforms the state-of-the-art ensembles in highly imbalanced data-sets.•EUSBoost behavior is explained via kappa-AUC error diagrams.
论文关键词：Classification,Imbalanced data-sets,Ensembles,Class distribution,Kappa-error diagrams,Boosting
论文评审过程：Received 14 December 2012, Revised 21 March 2013, Accepted 9 May 2013, Available online 15 May 2013.
论文官网地址：https://doi.org/10.1016/j.patcog.2013.05.006
