Partial Least Squares Regression

[#1]

Hi,

I have a few questions about the k-fold cross-validation method in JMP for partial least squares regression.  I would really appreciate any help you can provide.

1. Are the results from the folds averaged to give the Press statistic, or is the model with the best validation statistic chosen as the final model?  If the latter, would the Press statistic from leave-one-out cross-validation be more representative of the data?
2. Is the "root mean Press" the same as the RMSEP (sqrt(Press/n))?

Thank you!

[#2 · Accepted solution]

I emailed JMP Technical Support these questions and they replied as follows:

1. Yes, the RMPRESS statistic is calculated from the results from different folds.  And then the model with the smallest values of RMPRESS is chosen as the best model.
2. RMPRESS is the square root of the average squared residual from this prediction.

Hopefully this will help anyone else with similar questions.