Beyond Machine Learning Models, Time Series Forecasting plays an integral role for bussiness depending upon outlets and sales. So I tried an approch of solving a most common day to day customer visits to outlets, whose prediction can be bettered using both Linear Regression and Time Series Forecasting.
-
About 20-25% of the values are zero for customers which may be an indication of store closure for that day.
-
Impact of the promotion is clearly visible in the grouped boxplots below even across month. However for December, no impact of promotion is visible.
-
Customer count was high when there is a school holiday.
-
Dickey-Fulher tests suggested stationarity of time-series.
-
Various models are built with and without considering seasonality and a SARIMAX model with Time Varying Linear Model performed the best according to rmse values.
