A bias or systematic error is quite common when monitoring predictions vs reference data. Anyway we must have certain control limits to decide if the Bias is significant or not.
Procedures (as for example ISO 12099 )give details about how to calculate the Bias Control Limits (BCL). The idea is a "T test" to calculate if the differences between the mean predicted values mean and the mean reference values are significant different from cero.
Procedures (as for example ISO 12099 )give details about how to calculate the Bias Control Limits (BCL). The idea is a "T test" to calculate if the differences between the mean predicted values mean and the mean reference values are significant different from cero.
This limits will be a certain percentage of the SEP (Standard Error of Prediction).
We can add all this calculations into R and improve the Monitor function to receive a warning if the adjustment of the Bias is or not necessary.
I record this video: Should I adjust the Bias? to see how fast is R to run these monitor, and how we can customize the plots the way we like.
Function can be expanded with more warnings, like (for example) if it is necessary to adjust the slope.
I record this video: Should I adjust the Bias? to see how fast is R to run these monitor, and how we can customize the plots the way we like.
Function can be expanded with more warnings, like (for example) if it is necessary to adjust the slope.
No hay comentarios:
Publicar un comentario