I have data where an observation was made in 10 minute intervals for 8 weeks. I have around 170 variables that were measured every 10 minutes. I am trying to use multivariate time series analysis to predict what will happen in the 9th week (also in 10 minute intervals). I know that the ARIMA model is useful for these cases. But I am very new to R and statistics and I am having a little trouble starting out. Most of the information and tutorials on R that I found online are regarding single variate time series analysis and yearly/quarterly measurements, and not every 10 minute measurements, so it is difficult to apply to my problem. I was hoping to get some help on here and I would greatly appreciate any advice !
Not sure if this will help, but your problem reminds me a little bit of macroeconomic time series modelling where a similarly high number of regressors is available, although the time series are much shorter. J. Stock and M. Watson have a nice overview paper on dynamic factor models employed in this kind of setting.
There you should find some tips on how to reduce the number of variables in your system to a comfortable level. This would allow you to model them as a vector autoregression (VAR) or similar.
- Stock, James H., and Mark W. Watson. "Dynamic factor models." Oxford Handbook of Economic Forecasting. Oxford University Press, USA (2011): 35-59.
This may be of help:
You could do this as a multiple regression problem.
Depending on which variable you want to predict, you could introduce a time variable along with lagged predictors. For each of the lag of lagged predictors, you could determine which is the most correlated (be sure to use cross validation to avoid data dredging). Then include dummy variables based on the hour, weekday. See https://www.otexts.org/fpp/5/2
By this point, you will need to weed out the excessive variables so use a cross validated feature selection approach.