# How to conclude from RMSE and R-suared value that our model is good or bad?

I used two different columns from dataset as targets and using logistic regression.

Output for target 1

The model performance for training set
--------------------------------------
RMSE is 8.136996958218045
R2 score is 0.5261727899306442

The model performance for testing set
--------------------------------------
RMSE is 8.11008545942948
R2 score is 0.5156169844890891


Output for target 2

The model performance for training set
--------------------------------------
RMSE is 1.6402023883989456
R2 score is 0.5242729273650613

The model performance for testing set
--------------------------------------
RMSE is 1.4717139217150437


As am splitting data into 85:15 split getting these values as required.Before further proceeding to extract PCA components , I want to be confident that its a good fit.

My question is What to conclude from the given values and is there any pre determind good or bad value for RMSE and R2?

• The trouble with something like this is that it is tempting to use something like $R^2$ like a letter grade in school, where $0.9$ is an A that makes us happy, and $0.6$ is an F that makes us sad. However, $R^2=0.6$ could be quite excellent performance in some cases, while $R^2=0.9$ could be quite pedestrian in other cases.
– Dave
Aug 15 at 12:41
• @Dave that mean making any judgment on the basis of R2 could be dicey. Aug 15 at 13:06
• You just have to put it in context. // Additionally, $R^2$ even loses its usual “proportion of variance explained” interpretation in nonlinear models. It’s as valid of a performance metric as MSE or RMSE (just look at the equation), but it does lose that interpretation in nonlinear models, which I think diminishes its utility.
– Dave
Aug 15 at 13:11
• @Dave i suppose may be its better to use 'chi-square test' then . Aug 15 at 13:42
• What would that tell you? // Which chi-squared test do you mean? // I still say that the context is what matters.
– Dave
Aug 15 at 13:43