There are 50 shops and each shop have 30000 items. Goal is to forecast the sale of item based on shop.

enter image description here

Forecase the item_cnt_day, for this i dont see this as multi variate time series. Only shop and item ID is needed to forecast the next month data.

The question is do we need to take this as Multi time series problem and build 30000 ARMA, ARIMA, SARIMA etc models for each of the shop and items. So for 50 shops and 30000 items, there will be 50*30000 models.

What do you guys suggest to handle this problem as time series problem


1 Answer 1


The approach you suggested works, but with the sheer amount of time-series you have, plus having data dating back to 2013; I think it would be a good idea to try to create a single model trained on multiple time-series. This is easily done using the Darts Module.

Darts has examples you can follow for multiple time-series here.

  • $\begingroup$ I have data on two years daily sales. And looking to forecast monthly sales. Approach is to forecast dailysales for next month and then total to make it as monthly sale. $\endgroup$
    – Shan Khan
    Commented Dec 10, 2021 at 5:08

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.