1
$\begingroup$

I try to predict a score $S$ at the end of a time $Τ$ with measure taken from time $h_1$ to $h_f$ (~10 constant intervals) from different sources (always different but close behavior).

My goal is to try to predict $S$ at time $h_1$, $h_2$, ... , $h_f$.

My measure evolve during $h_1$ to $h_f$ but $S$ stay the same. I don't know if I can consider this dataset as time series because target doesn't change during time and measures are taken from different sources. Apply an ARIMA model could be a good idea ?

$\endgroup$
4
  • $\begingroup$ Are the lengths of your time series fixed? $\endgroup$
    – JB1
    Jun 29, 2018 at 11:47
  • $\begingroup$ yes, intervales between two h is constant $\endgroup$ Jun 29, 2018 at 12:17
  • $\begingroup$ but how long are each of the sequences you want to analyse? or do you want to look at individual time points= $\endgroup$
    – JB1
    Jun 29, 2018 at 12:30
  • $\begingroup$ ah sequence are also constant, each instances have the same number of measures at the same intervals. $\endgroup$ Jun 29, 2018 at 12:35

1 Answer 1

0
$\begingroup$

depending on the use case, exactly what you are trying to achieve and the outcome there are a few methods you could use to approach this problem.

Mixed effects models allow you to model fixed and random effects (time and measurement) with respect to some outcomes and other random effects (age etc).You can then use this as a predictive model to predict in unseen data.

In a more "machine learning" approach, LSTMs/RNNs can use sequence analysis to predict an outcome or for forecasting. These (generally) require that your sequences are of equal length.

$\endgroup$
2
  • $\begingroup$ I don't know if i can treat my data as regular times series because each points in a sequence have the same target and features are time-dependant. $\endgroup$ Jun 29, 2018 at 13:03
  • $\begingroup$ For LSTM/RNN , the shadow part is what should be a batch ? Sequence should be mixed or 1 batch = 1 sequence ? $\endgroup$ Jun 29, 2018 at 13:04

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.