Ideal values for NRMSE and Decomp.RSSD

[CJ2407]

Hi,

Based on your experience, I would like to know your thoughts on the following to be able to understand when to conclude my modeling exercise with more confidence -

What are ideal values for NRMSE for Validation and Test datasets?

• Mine is a retail company
• R2 train = 85%, R2 Val and R2 Test = 73%
• NRMSE train = 0.0301, NRMSE Val and Test = 0.65 to 0.75

Till what point should we continue running iterations for Decomp.RSSD to converge? My model almost converge on NRMSE, but almost never on Decomp.RSSD.

• current decomp.RSSD value is 0.09. How to interpret this value? What's the goal range for this metric?