I am fitting a Lasso Regression to do feature selection in my dataset. I have seen it is common practice to use StandardScaler to standardize the dataset. However, given that the distribution of my dataset is non-Gaussian, does it still make sense to use StandardScaler, if so, why? Or are other standardization techniques preferred?
$\begingroup$
$\endgroup$
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
2
It is generally recommended to standardize the dataset before fitting a Lasso regression model, regardless of the distribution of the data. This is because standardization ensures that all features are on the same scale, which can improve the performance of the Lasso regression model.
Standardizing the data allows the Lasso regression model to put equal emphasis on all features, which can improve the model's ability to select important features and reduce the magnitude of the model coefficients.
answered Dec 8, 2022 at 9:51
-
$\begingroup$ Thank you. But I've read that StandardScaler is used if we assume the data distribution is normal. Should we use a different method to standardize should that not be the case (such as z-score)? $\endgroup$– StephMCommented Dec 8, 2022 at 10:15
-
$\begingroup$ Other standardization techniques may also be used, but StandardScaler is a widely used and effective technique that is well-suited for use with Lasso regression. $\endgroup$ Commented Dec 8, 2022 at 10:38
$\begingroup$
$\endgroup$
Normalization is recommended with penalized methods such as LASSO, however the choice of normalization is up to you, you could use standardization, min-max normalization, or any other of the many available normalizations, LASSO does not care.
Your variables do not have to be normally distributed (with most models) and it is not true that you can only use standardization on a normally distributed variable, it will work on any variable, that is it will achieve a mean of zero and SD of one.
