0
$\begingroup$

Could any one help me know about different approaches, methods or algorithms to build a model to forcast a time serie which has only two values ( 0, 1 ) but they last over time each time . basically I've some on/off data, it tell me if there is an object in a place and how much time that object was in that specific place . So the output is a rectangular wave but it's not periodic.

so if we have the object in that place field we will get 1 every x seconds then when there is nothing we will get 0 . Thanks in advance of your response!

$\endgroup$
2
  • $\begingroup$ I would model it as a sequence of on and off events, using a kind of Poisson process and trying to infer its parameters. $\endgroup$ Jul 17 '18 at 1:48
  • $\begingroup$ I don't think Poisson Process will work in this case since it mostly used to model rare events such as machine malfunction ... $\endgroup$ Jul 17 '18 at 9:21
0
$\begingroup$

You could just predict if the next hour a room will be occupied based on whatever features you give it. Each Example would be per room per hour. You can then use whatever machine learning model you want. It saves needing to go to the complexity of some sequence models - could just use Logistic Regression, for example.

Some examples of features:

  • Was it occupied in the previous hour?
  • How many of the previous hours was it occupied for?
  • Of the past 10 hours, how many were it occupied for?
  • Since the start of the 'working' day, how many hours has the room been occupied for?
  • Was it occupied the same hour yesterday? Was it occupied the same hour a week ago?

You could obviously add any others and incorporate extra information, if you have it.

$\endgroup$
0
$\begingroup$

First you should define your goal for the prediction. Do you want to predict also a time series output? Then the best thing to do is to use any machine learning model that is sequence focused (e.g. is LSTM), and with the time series that you have as input. Also study the time scale of your model. Study if it should be hour, minutes or any other scale.

$\endgroup$

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