What to do after fitting regression model on scaled data?

Question

I have created a multiple-linear regression model on some data. The different explanatory variables were all in different units, so it seemed appropriate to scale the data. I've fitted a model and the fit seems decent, however I'm not too sure what to do now. I want to talk about what the coefficients show and discuss the size of the residuals, but that would be with respect to scaled data, rather than the original data. What can I do to go from having a 'scaled' model to having an 'unscaled' model?

...it seemed appropriate to scale the data... in that case you explain the model on the scaled data. If you do not want that then do not scale your data. — user2974951, Jan 10 '24 at 11:50
How can I relate my thoughts on the scaled data to the non-scaled data? — Ewan McGregor, Jan 10 '24 at 11:51
Did you perhaps use a ridge or LASSO penalty? Otherwise you'll get an equivalent model whether you scale the predictors or not; & in any case the coefficients can simply be transformed back for any linear transformation. — Scortchi - Reinstate Monica, Jan 10 '24 at 12:11
@user2974951 So you're telling me there is no way to convert the model coefficients to be 'unscaled'? — Ewan McGregor, Jan 10 '24 at 13:07
I am telling you that unless you have good reasons (and different scales are not it) to standardize the data then this is a pointless exercise (and if you are going to standardize the data then you will probably want to interpret them on a standardized scale). — user2974951, Jan 10 '24 at 13:22
@user2974951 That's ridiculous, data having different scales is probably the most common reason to scale them. If your variables have different units then any machine learning algorithm won't return accurate results. — Ewan McGregor, Jan 10 '24 at 13:59
When conducting multiple regression, when should you center your predictor variables & when should you standardize them? — user2974951, Jan 10 '24 at 14:03
@EwanMcGregor (Generalized) Linear models don’t care if you scale the data. For deep learning, there are computational advantages when it comes to the numerical optimization, sure, but even the classical theorems make no mention of scaled data. — Dave, Jan 10 '24 at 14:03

What to do after fitting regression model on scaled data?

0 Answers0