1
$\begingroup$

I have a small dataset which has timestamp and temperature values for 6 months(I.e. one temperature value per day). I would like to forecast 2-3 months of temperature. I would like to know, what kind of models would be suitable for such purpose.

I have used LSTM but the forecasts degrades as I go further and the data size is pretty small for LSTM to be trained.

$\endgroup$
  • $\begingroup$ for LSTM to work well, you need lots of data. Did you try ARIMA, AR, OR VAR models? $\endgroup$ – i.n.n.m Nov 30 '17 at 22:11
1
$\begingroup$

You can use reinforcement learning, in instance mbrrl2 package.

As temperature is very stational dependent variable. You should have at least 1 year, better 2 years, to make some predictions.

Other way to solve this, is having a collection of historical temperatures, clustering them, and matching your dataset with the closest cluster.

| improve this answer | |
$\endgroup$

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.