Open Swapnilpawar29 opened 2 years ago
Hey, you might have better luck throwing away the data from before October 2020 (when the spikes get bigger and there's a level shift in the series). This will disable yearly seasonality, but there doesn't seem to be much of that in this time series (it's moreso regularly spikes, weekly seasonality, and maybe some monthly seasonality it seems).
Hello Folks,
Any help would be appreciated. I am trying to build a 12 future months forecasting model based on limited daily data that's available from Jan 2020 up to current month. Could someone help me with this simple question. Is this time series suitable for fbprophet application. My problem is no matter how many different ways I try playing with different parameters, the model keeps on building forecast below the observed trend.
P.S. - I have managed to handle spikes by passing them as holidays. Also future months are not shown in the second graph below. Original series:
Original and Forecast together: