On the appropriateness of Platt scaling in classifier calibration
作者：
Highlights：
• Proof of the general parametric assumptions of Platt scaling.
• Correcting incorrect statements from related work.
• Equivalence proof of Platt scaling and beta calibration, up to a preprocessing.
• Analyzing evaluation metrics and showing that popular ones yield suboptimal results.
• Supporting the theoretical findings in a simulation study with perfect information.
摘要
•Proof of the general parametric assumptions of Platt scaling.•Correcting incorrect statements from related work.•Equivalence proof of Platt scaling and beta calibration, up to a preprocessing.•Analyzing evaluation metrics and showing that popular ones yield suboptimal results.•Supporting the theoretical findings in a simulation study with perfect information.
论文关键词：Platt scaling,Classifier calibration,Posterior probability,Logistic regression,Proper scoring rules
论文评审过程：Received 11 April 2019, Revised 3 July 2020, Accepted 21 August 2020, Available online 10 September 2020, Version of Record 11 September 2020.
论文官网地址：https://doi.org/10.1016/j.is.2020.101641
