Regularization application to features

Hello, i understand that regularization needs to be applied to the features of a linear or logistic regression model, in order to ensure that all the feature values fall more or less within a small range of values.
My question is regarding the application of regularization. Is it necessary to apply it to all the features of a model, or only to specific features?
If we only apply regularization to specific features, does it change or alter the relationship between input features and output and hence result in the model getting “corrupted”?

No, that’s not what regularization does.

What you described is feature normalization.

Regularization is used to control overfitting, by reducing the magnitude of the weights.
Yes, it is applied to all features.

1 Like

If we apply regularization selectively to some features, we would be altering the impact of these features in the overall equation - This would most definitely have an impact on the predictive power of the model.

Apologies for using the wrong term. What i meant was normalization only.
Can we selectively apply normalization to only those features displaying a much larger range of values and leave other features alone?

Yes, you could do this if you wish. But it makes for rather complicated and customized code for each data set.