A unified view of class-selection with probabilistic classifiers

作者：

Highlights：

• A unified framework of class-selection by the use of probabilistic equivalence is presented.

• A learning procedure for adapting the probabilistic equivalence is proposed.

• A normalized performance measure for class-selection is described.

• Experiments on real world datasets, using different classifiers, show the (statistical) significance of the proposition.

摘要

•A unified framework of class-selection by the use of probabilistic equivalence is presented.•A learning procedure for adapting the probabilistic equivalence is proposed.•A normalized performance measure for class-selection is described.•Experiments on real world datasets, using different classifiers, show the (statistical) significance of the proposition.

论文关键词：Reject options,Multiple class-selection,Probabilistic classification,Probabilistic metric

论文评审过程：Received 12 December 2012, Revised 27 May 2013, Accepted 29 July 2013, Available online 7 August 2013.

论文官网地址：https://doi.org/10.1016/j.patcog.2013.07.020