I have 6 cycles of historical time series data of retail sales and using a prophet model to forecast one cycle. I am using 4 variables as regressors. There is one month in particular - 'April', which is being under predicted by the model repeatedly, the sales dip drastically in that month and that has been the case for all the cycles. The predictions are giving around an average of 25% absolute error.
I have tried:
- To add 'month' calendar variable
- To add a dummy regressors with April flagged as one and others as 0
- To add lagged column of the retail sales
- To add decomposed seasonality of retail sales
as additional regressors, trying out various combinations of them all. But the model is still not being able to capture 'April' months sales.
Any other options I have?? Would really appreciate any help.