0
$\begingroup$

In dealing with uneven spaced time series data, any advise what would be the approach ? data is ECG data to predict if the blood pressure Sys would drop -20% or 80% of normal. In the usual approach for predicting weather we have regulared spaced time series data, how would you advise the approach for this situation in unevenly spaced time data.

Here is a sample look of the data uneven spaced time data

$\endgroup$
2
  • $\begingroup$ Is Duration the time between measurements? $\endgroup$
    – m13op22
    Commented Mar 26 at 13:41
  • $\begingroup$ yes Duration the time between measurements, the data is a heartbeat dataset, duration and IBI is somewhat the same @m13op22 $\endgroup$ Commented Apr 1 at 5:14

1 Answer 1

2
$\begingroup$

It depends what you want to do with it later. Certainly the easiest thing would be to do interpolation of some sort and resample the time series at evenly spaced points.

More complex would be to use evenly spaced points as a latent model, and then have interpolation from that sampling into your un-even sampling as an extra step. You could then estimate latent parameters using something like maximum likelihood.

Another complex way would be to use some continuous model, either biology/physics - inspired or a generic Gaussian process, and then add sampling into discrete uneven time-points as a part of data acquisition modelling.

Yet another method would be to re-derive ARIMA for un-evenly spaced spaced points. This will sort of be similar to the previous method. The down-side will be that you will have to re-check all typical ARIMA results to select which ones apply to you. I am sure someone would have done it, but I don't know of such literature

$\endgroup$
3
  • $\begingroup$ Good answer! But for more detail, what's the benefit of doing the complex methods over the easiest step? $\endgroup$
    – m13op22
    Commented Apr 1 at 14:05
  • $\begingroup$ @m13op22, if you are going for interpolation, then you are obscuring your data generation process. it becomes difficult to write the likelihood of your data under some distributional assumptions. this will make it difficult to, for example, evaluate the credible range of your inferences. $\endgroup$
    – Cryo
    Commented Apr 2 at 20:01
  • $\begingroup$ one way to look at it is that interpolation is a step that extracts a statistic from your raw data. Statistic could be inefficient in terms of providing inference results, or could bias your inferences. Establishing all the necessary guard rails around such statistic would negate all apparent ease. Or you could ignore the guard rails… might just work in this case $\endgroup$
    – Cryo
    Commented Apr 2 at 20:06

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.