F-measure curves: A tool to visualize classifier performance under imbalance

作者：

Highlights：

• The F-measure space is proposed to visualize classifier performance.

• The preference between precision and recall is controlled by a factor.

• A crisp classifier is represented as a curve for a range of P(+).

• A soft classifier is shown as the upper envelope of curves of its decision thresholds

• The proposed Iterative Boolean Combination(IBC) algorithm is optimized in this space.

• The proposed IBC selects and combines classifiers in this space.

摘要

•The F-measure space is proposed to visualize classifier performance.•The preference between precision and recall is controlled by a factor.•A crisp classifier is represented as a curve for a range of P(+).•A soft classifier is shown as the upper envelope of curves of its decision thresholds•The proposed Iterative Boolean Combination(IBC) algorithm is optimized in this space.•The proposed IBC selects and combines classifiers in this space.

论文关键词：Pattern classification,Class imbalance,Performance metrics,F-measure,Visualization tools,Video face recognition

论文评审过程：Received 11 May 2019, Revised 30 October 2019, Accepted 27 November 2019, Available online 16 December 2019, Version of Record 29 December 2019.

论文官网地址：https://doi.org/10.1016/j.patcog.2019.107146