I am trying to make a time series forecasting model for a 11000 customers to predict their individual turnover in the next 30 days. The data given to be is on a daily basis for each customer. the dataset has the following columns- Date, customer id, turnover, place
The issue is when the customer does not make any transaction, the turnover for that date is zero and for every customer I end up having a lot of zeroes which makes my data imbalanced.
How to deal with this sort of data to make machine learning models for forecasting?
I have tried Arima and Prophet. But none of them seem to give decent predictions because of the number of zeroes in the data.
Also, can I do feature engineering and make new features? If yes, what features would you recommend?