Geometric one-class classifiers using hyper-rectangles for knowledge extraction
作者：
Highlights：
• We propose two efficient geometric one-class classifiers using hyper-rectangles.
• They can control the scale of the h-rtgls for the accuracy and overfitting issue.
• They can provide interpretability on the classification results.
• They work well on complex datasets such as banana-shaped datasets.
• We show the superiority of the proposed classifiers by a numerical experiment.
摘要
•We propose two efficient geometric one-class classifiers using hyper-rectangles.•They can control the scale of the h-rtgls for the accuracy and overfitting issue.•They can provide interpretability on the classification results.•They work well on complex datasets such as banana-shaped datasets.•We show the superiority of the proposed classifiers by a numerical experiment.
论文关键词：One-class classification,Hyper-rectangle,Feature space,Interval merging,Interval partitioning,AUC,Interval conjunction,Interpretability,Rule extraction
论文评审过程：Received 22 January 2018, Revised 27 August 2018, Accepted 19 September 2018, Available online 20 September 2018, Version of Record 29 September 2018.
论文官网地址：https://doi.org/10.1016/j.eswa.2018.09.042
