Numerical likelihood ratios outputted by LR systems are often based on extrapolation: When to stop extrapolating?

详细信息    查看全文

• 作者：Peter Vergeer ; ^{p.vergeer@nfi.minvenj.nl} ; Andrew van Es ^{a.van.es@nfi.minvenj.nl} ; Arent de Jongh ^{a.de.jongh@nfi.minvenj.nl} ; Ivo Alberink ^{i.alberink@nfi.minvenj.nl} ; Reinoud Stoel ^{r.stoel@nfi.minvenj.nl}
• 关键词：Strength of evidence ; Validation ; Calibration ; Normalized Bayes error ; Likelihood ratio
• 刊名：Science & Justice
• 出版年：2016
• 出版时间：December 2016
• 年：2016
• 卷：56
• 期：6
• 页码：482-491
• 全文大小：781 K
• 卷排序：56

文摘

Numerical likelihood ratios outputted by LR systems are often based on extrapolation. LR systems should be made robust to extrapolation errors. Robustness is achieved by setting a minimum and maximum value to the LR. These values may be determined with the Empirical Lower and Upper Bound LR method.