2
$\begingroup$

I understand that feature scaling is required to bring features in different magnitudes on a common scale so the model is not biased towards features with higher magnitudes. But if there is only a single feature in a feature set. For example, I have a quantity (events over time) on a time series and I'm forecasting for the future. In this case, do I need feature scaling (normalization or standardization?) or will it be unnecessary?

I searched online but couldn't find resources talking about the case where there's only a single feature in the feature set.

$\endgroup$
1
  • $\begingroup$ The entity is the same but I believe you must be using its value at different intervals that as multiple Features. So you should. $\endgroup$ – 10xAI Nov 26 '20 at 16:21
1
$\begingroup$

I understand that feature scaling is required to bring features in different magnitudes on a common scale so the model is not biased towards features with higher magnitudes. But if there is only a single feature in a feature set. For example, I have a quantity (events over time) on a time series and I'm forecasting for the future. In this case, do I need feature scaling (normalization or standardization?) or will it be unnecessary?

Usually we would try to do feature scalling although the idea is to remove the dependence of absolute magnitiude in affecting convergence. it will help if the distribution is made gaussian with zero mean and unit variance. Since the gradient will be lower the stochastic descent will be a lot smoother.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.