3
$\begingroup$

If I have a list of values throughout time, say a list of values for every minute throughout an hour of monitoring something, can I somehow 'predict' or estimate what the value would probably be in the future, say another 20-30 minute after my last known value? And if so, how can this be done?

I figure maybe I could do it with just a 'best-fit' to a quadratic and look at the value for x=90 (60 known values + 30 values into the future) but I feel like that probably isn't going to be the most accurate method. Does anyone have any suggestions for a better way to do this? Just a note, I do have very little statistical experience, so please bare with me if this question is known to be easy (or impossible).

Thanks!

Improve this question
$\endgroup$
3
  • $\begingroup$ This is a very general question that no one can answer without more information. It really depends on the nature of your data. Do you have a reason to believe that past behavior can predict future behavior? Do you have other variables you can use to help make predictions (for example, if you are trying to predict future temperature, knowing the pressure could help)? $\endgroup$
    – rm999
    Commented Apr 29, 2011 at 17:57
  • $\begingroup$ Hm. The nature of the data tends to be mostly oscillating, but not entirely and not always. Sometimes it tends to be more linear. No, I don't really have any other variables that could be applied. Predicting this could just be too difficult... $\endgroup$
    – JToland
    Commented Apr 29, 2011 at 18:15
  • $\begingroup$ Hey! did you find an answer to your question? I have the same doubt. $\endgroup$
    – LoveMeow
    Commented Oct 8, 2014 at 15:33

2 Answers 2

Reset to default
2
$\begingroup$

You might try using ARIMA models making sure that you incorporate any identifiable Level Shifts and/or Local Time Trends culminating in an ARMAX model. Changes in parameters/variance of the errors should also be tested and remedied if necessary. As compared to NN, these approaches challenge the data rather than simply believing/fitting the data.

Improve this answer
$\endgroup$
1
$\begingroup$

Yes, the problem occurs in variety of scenarios for time series prediction thing. But I would suggest to avoid going through quadratic analysis just by estimating that it might fit.

Go on hitting with Linear multivariate regression, further try a combinations of non-linear series. Plus I have also seen Certain problems of this sort like retweet prediction, stock prices predictions giving nearly accurate results with SOFNN- self of organizing fuzzy and neural network. Try scholar.google search, you would hit upon a previous good work on this.

Improve this answer
$\endgroup$

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.