I am working with variable length time-series signals. I want to use a sliding window to extract features, things like mean, standard deviation, kurtosis, skewness. The length varies pretty significantly.

However for the training I will be using linear models like Logistic Regression and I need a constant input length.

I'm not sure how to equate the signals, should I use a fixed-size window and use the largest signal input length while padding the smaller signal features with zeros?

Should I use a variable window-size and divide the signals into a fixed predetermined number of segments?

I'd really appreciate some suggestions. Sorry if this is a duplicate, I could not find the answer, perhaps because I was searching for the wrong thing.

  • $\begingroup$ It depends on the problem you try to solve. I.e. what is the distribution of length of sequences? Can you wait for the full sequence before predictions or do you need to do it "in flight" when any new signal is comming, etc. $\endgroup$ – wind Apr 23 at 11:42

Your Answer

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

Browse other questions tagged or ask your own question.