Does Data Normalization affect RMSE & MAE?

Question

Suppose I am using two different methods to perform regression, such as ARIMA (1) and a Neural Network (2), and I am using the RMSE and MAE metrics to measure accuracy. For method (1), I do not normalize the data, while for method (2) I do.

When I compute the accuracy metrics, will these be scaled differently, due to normalization? If so, do they need to be scaled back?

Thanks in advance.

Answer 1

Ideally, RMSE and MAE is measured in the same units of your target variable. If the normalisation has effect in your prediction ie Improves/deteriorate this might affect the RMSE and MAE scores but not the way it is scaled. Hope this answers your question.