About standardization/normalization

[sunjar2020]

Dar Sir:

I am using tsfresh to try to extract features from stock data. My data contains the common OHLC price, Volume and several of my own factor columns, which are recalculated data based on fund activity. My question is, do I need to standardize/normalize my data before feature extraction?

When I used LSTM to make time series forecasts before, because the OHLC of stocks and their price moving averages are usually several orders of magnitude smaller than factors such as volume and funding, I group them and then perform standardization/normalization. But what I don't understand is whether I need to do this when using tsfresh? I know that tsfresh itself will extract some statistical features (such as mean, standard deviation, etc.). Will these features lose their meaning after standardization? So should I use tsfresh directly to process this data, or do I need to standardize/normalize it?

[nils-braun]

Hi @sunjar2020 !
Sorry for the delayed answer.
I think the answer to your question depends on what you want to do with the extracted features afterwards.
If you want to use them to understand your data better, you probably do not want to normalize them. If you want to use them for feeding into a ML model, you might want to normalize your data - depending on the ML model you use. Please note however that also the result of the extracted features might need a normalization again, depending on the ML model.

If you are really trying to train a model from it,I would actually suggest to try out all possibilities and see which performs best on your data and model. Just make sure you apply the same normalization for training, validation and test sets.

[sunjar2020]

Thank you very much for your reply, you make sense, the best way is to try both situations, got it.