A hybrid data-level ensemble to enable learning from highly imbalanced dataset

Zhi Chen, Jiang Duan, Li Kang, Guoping Qiu

Abstract: Highlights•An ensemble method for learning from highly and extreme highly imbalanced datasets is proposed.•A new margin-based undersampling scheme is designed to remove the less representative majority samples.•A diversity-enhancing oversampling method is proposed to generate high-quality minority samples.•Comprehensive experiments have been conducted to prove the effectiveness of proposed method prove.