After performing ordinary kriging on my dataset using gstat package in R , I have done 10-fold cross validation to determine some statistical parameters (R,RMSE, MB etc.) Due to procedure, 10- fold cross validation give different value of R/ RMSE each time I run the same code as its working by selection random fold.
suppose, first time, RMSE = 17.5, second time, RMSE =17.4, third time, RMSE = 17.55...and so on. So, In this case, which value of RMSE should I take as final RMSE value? should I take the range of RMSE value? but generally in scientific paper, I saw the R/RMSE value is show as single number not as range.
Thanks in advance.
Uzzal
I would suggest taking the mean value of RMSE across all kfolds and then also including the 95% confidence intervals.
95% confidence intervals can be calculated as follows: 1.96 * SE
standard error (SE) is calculated as the standard deviation of "x" divided by the square root of the sample size.
In R code: CI = 1.96 * (sd(x)/sqrt(length(x))
If you have access to "The New statistics for R" by Andy Hector I would highly recommend it. The main point of the book is to use confidence intervals for biological data instead of p-values.
- Badges
- Science topic