In retail sector few NOOS(never oot of stock items) exhibit a peculiar type of seasonality and mostly it was event base , and multiplicative in nature.Machine Learning models are very diffcilut to estimate that Multiplicative pattern YoY. So this arhictecture will build on Box-Jenkinson methodology,it works as follows:
- Identfy the series is Stationary or Non-Stationary
- Tag the treatment require for make it a stationary time series
- Then it will find for Yearly seasonality Exist or Not
- Then do the simulation using auto_arima() -- both the parameters-p,d,q & P,D,Q
- Then it will do the forecasting