Optimizing area under the ROC curve via extreme learning machines

作者：

Highlights：

• We propose a novel off-line binary AUC optimization algorithm called ELMAUC by bridging the least square AUC optimization method with ELM framework.

• Two potential multi-class extensions of AUC are compared theoretically.

• A unified objective function for multi-class AUC optimization is proposed. Subsequently, two novel off-line algorithms named ELMMAUC and ELMmacroAUC respectively are proposed for multi-class AUC optimization.

• The generalization analysis of ELMMAUC is established.

• The experimental results on 11 binary-class datasets and 15 multi-class datasets suffering from class imbalance problem show the effectiveness of our work.

摘要

•We propose a novel off-line binary AUC optimization algorithm called ELMAUC by bridging the least square AUC optimization method with ELM framework.•Two potential multi-class extensions of AUC are compared theoretically.•A unified objective function for multi-class AUC optimization is proposed. Subsequently, two novel off-line algorithms named ELMMAUC and ELMmacroAUC respectively are proposed for multi-class AUC optimization.•The generalization analysis of ELMMAUC is established.•The experimental results on 11 binary-class datasets and 15 multi-class datasets suffering from class imbalance problem show the effectiveness of our work.

论文关键词：Extreme learning machine (ELM),Area under the ROC curve (AUC),Imbalanced datasets,Multi-class AUC optimization

论文评审过程：Received 23 November 2016, Revised 15 May 2017, Accepted 17 May 2017, Available online 18 May 2017, Version of Record 6 June 2017.

论文官网地址：https://doi.org/10.1016/j.knosys.2017.05.013