How to compare the accuracy of linear regression when normalized in different ways?

I looked up various models for regression.

Using Matlab, there were several models such as SVM, linear, and Gaussian.

I found normalization and PCA help train the data.

I trained the data using various normalization methods in Matlab.

How do I know which normalization method works better?

If it is immediately evaluated using MAE or MSE, a small-scale normalization method would be advantageous.

For example, the normalization method of the range [0 1] will have a lower MSE than the normalization method of the range [-1 1].

Should inverse functions be used for evaluation to return to the range that the original data had?

Answer 1

[Angelo Yeo]

Hi 영훈 정,

You need to reverse the normalization in order to assess the performance of regression. This is because, as you mentioned, different normalization results in different performance result.

Also, FYI, when you choose which normalization method to use, you cautiously take a look at the property of dataset. This can be either the distribution of the data or the existence of outliers (or noise).

Hope this helps,

Angelo