AUC-Based Extreme Learning Machines for Supervised and Semi-Supervised Imbalanced Classification

Extreme learning machines (ELMs) has been theoretically and experimentally proved to achieve promising performance at a fast speed for supervised classification tasks. However, it does not perform well on imbalanced binary tasks tends get biased toward the majority class. Besides, since large amount of training data with labels are always available in real world, there is an urgent demand develop efficient semi-supervised version ELM In this article, owing distinct insensitivity area under ROC curve (AUC) both class skews changes distributions, we focus study integrating AUC maximization into framework tackle well. By demystifying metric framework, new AUC-based called AUC-ELM classification, which essentially revealed be equivalent another transformed space. Accordingly, its SAUC-ELM also developed. Both have distinctive merits: 1) they share advantage generalization capability efficiency, further uniquely tailored 2) contrast existing variants ELM, such as class-specific cost regulation fewer parameters tune, thereby reducing computational model selection. Experiments heap datasets show that outperform other comparative methods terms speed.