Suppose I am observing the traffic over a network node every minute of a single day. At the end of the day, I have a table of observations like this:
Time | Traffic
00:00:00 | t1
00:00:01 | t2
. |
. |
. |
23:59:00 | tn
What kind of models are there to offer a prediction for the traffic on the next day based on these observations?
The naive approach I can see would be to choose a time interval, say an hour, take an average of the traffic over that hour and use it as my prediction for every minute of the corresponding hour the next day.
(a) Is there anything else you can think of in terms of prediction with this data?
(b) I am also know considering validation. How would you validate this model? Say a model produces predictions, what validation techniques are there given only this dataset? I am thinking cutting it up in between time intervals, so say 16 hours spread throughout the day are chosen to run/train the algorithm, and the remaining hours interspersed throughout the day can be used to validate.