I have a daily time series, with weekly seasonality, but it starts and ends in the middle of the week (Wed/Thu, during the season peak).

I have two questions:

  1. Should I drop the first and last partial weeks from the data before modelling?

  2. When splitting into training and test sets, should I also make sure to include only whole seasons in each set?


1 Answer 1

  1. No, it is probably better to use and keep all data. Repeated-measures methods as implemented in functions like lmer from R package lme4, and gam and s from mgcv do not require higher-level units to have the same number of lower-level units.

  2. Yes. Cross-validated accuracy measures are likely less optimistic when you sample at the higher than at the lower level.

  • $\begingroup$ Thank you @Marjolein. $\endgroup$
    – Kazumi
    Feb 28, 2021 at 0:27
  • $\begingroup$ @Kazumi You are welcome $\endgroup$ Feb 28, 2021 at 9:34

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.