I have been studying about data scaling. Two common methods for it are the StandardScaler and MinMaxScaler. As I understood, StandardScaler expects the data to be normally distributed, but I have seem some examples online where this method is used for skewed distributions, so is it really important for the data to be normal in order to use standardization? And, if the distribution is important, can I use in the same dataset the MinMaxScaler for those features with skewed data and the StandardScaler for the features with normal distributed data? Would that be better than just choose one of the two methods and apply it on every feature?


1 Answer 1


Both are linear transformations. In general you should try both and see which performs better.

MinMaxScaler: Has the problem that your features will not have the same range of values after scaling. The advantage is that you might have intrinsic boundaries for your features. Also the interpretation of your variables ist still pretty straight forward.

StandardScaler: Your features will be on the same scale, which leads to better comparability between your variables. The Problem is that you do not know the mean and the standard deviation of the popluation but only of your training data set. Hence, you have to assume that these statistics are good aproximations for the population.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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