Interpretation of RMSE How is RMSE value interpreted? What makes it a good value?
I used the tidymodels collect_metrics() function and am getting an rmse value of 182 for one of my models. What does that mean?
 A: The RMSE (root mean square error) is, roughly, the standard deviation of residuals, and so measured in the same units as the response or outcome variable of a regression or similar model.
(Some people still say dependent variable.)
So it's 182 g, or 182 mm, or whatever it is in the units of measurement you have. (You tell us.)
It's loosely the typical size of difference between observed and predicted responses. Freedman, Pisani, Purves Statistics (any edition) is especially lucid explaining this.
You want it to be as small as possible, meaning at least two things:

*

*In comparing different models, lowest RMSE is best if other considerations are the same. (They usually aren't, but I can't helpfully summarize texts or courses on regression in one answer.)


*In your field, you should have some sense of what is big. If data are widely familiar, you should also have some such sense. For example, a model that predicts human female adult height with a RMSE of 10 cm sounds poor to me, while 1 cm is a lot better. (Often one benchmark is the kind of measurement precision you can expect. RMSE of about the same magnitude as measurement errors is often as about as good as you expect.)
Always keep a note of the sample size when you think about the RMSE.
Warning: Terminology varies and different texts and courses may use slightly different names for this beast.
Note: This answer pays no attention to the R code mentioned, which I have never used. If some detail in that software makes this answer incomplete or even incorrect in some sense, please sing out.
