The Use of Root Mean Square Error in Fit and Prediction

A different metric used to compare models with different ranges/standard deviations of the y value, is the root mean square error in prediction (RMSE/RMSEP). 

A lower RMSE would indicate a better model. The advantage of RMSE is that it expresses the residual error of the model in the original units of the dependent variable, and it can be directly compared to the experimental error in determining the quality of the prediction. As a quality measure, it is independent of the variance of the test compound dataset. It is a measure that can be used to compare different models, even if based on different datasets with different variances, unlike r and r.  

---