I understand that scaling data is important for certain machine learning algorithms, and the idea makes sense. I've found this great description of the processes here https://ourcodingclub.github.io/tutorials/data-scaling/#Scaling.

I have been running a hidden Markov model (using depmixs4) for an ecology question. How important is data scaling for HMMs? I have tried searching the literature and there are very few cases that I can find that data has been scaled prior to HMMs. None at all for ecology.

Am I missing something? Is data scaling not necessary for HMMs or can they cope with unscaled data? I can't really find any mentions of this in standard HMM guides.

For my model, I have four variables in different measurements, which are not normally distributed. I was planning on transforming data and then scaling appropriately. Papers that have used similar variables do not mention data scaling at all.

  • $\begingroup$ it's super implementation-specific. Unfortunately, it would not surprise me at all if scaling is not mentioned in a journal article that does it. In my view, it is almost always worth doing when using research code. $\endgroup$ Commented Jan 29 at 18:45


Your Answer

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