A possible classification for range of acceptability in the case of PLS regression is reported in Karoui et al., (2006) (Karoui, R., Mouazen, A. M., Dufour, É., Pillonel, L., Schaller, E., De Baerdemaeker, J., & Bosset, J. O. (2006). Chemical characterisation of European Emmental cheeses by near infrared spectroscopy using chemometric tools. International Dairy Journal, 16(10), 1211–1217.). Ranges are however based on R^2: values between 0.66 and 0.81 indicates approximate quantitative predictions, a value between 0.82 and 0.90 reveals good prediction, a value above 0.91 is considered to be excellent.

Regards,

Hello, the acceptable range varies from each type of sample and reference method. The acceptable range is yours to define. RMSEE and RMSECV are not unitless, they have the same units as your reference data, and provide an estimation of the mean error expected by your model in relation to your reference method.

To have a general idea of the range of RMSEE and RMSECV, try to read similar works of yours and compare the RMSEE and RMSECV of those works. As Marcello pointed out, there are other parameters that you should be looking into when creating a model, such as: bias, R squared, maximum and minimum error.

Hope this helps.

The Root Mean Squared Error of Estimation (RMSEE) is calculated as the root squared distance between the real Y variable - the estimated Y variable.

This means that its value depends on the original Y variable scale.

Now, if you are doing a classification metabolomics experiment (Control vs Treated) and say you get a 0.1 RMSEE, this means that generally your observation are predicted in average 9 times on 10 correctly.

The RMSEcv instead gives you an idea of the standard deviation of the RMSEE. If the RMSEcv is very different from the RMSEE, it means that every CV model is predicting worse your observation.

Now, in your shoes, I wouldn't care too much about it. RMSEE and RMSEcv are very important in non-binary classification/prediction problems, which I suppose is not your case (but if it is, please get me back to it).

Check The R2 and Q2 values as suggested by Marcello, the probability of the CV-ANOVA, and the misclassification table.

Good luck!!!