How can we use backtracking algorithm to improve the forecast accuracy in time series forecasting of products in retail industry. Can someone explain what should be the basis of backtracking…can it be MAPE?.
Hi @NISHIKET, could you give a resource where you saw the use of backtracking algorithm? Also could you elaborate on MAPE?
Thanks for replying back.
We are working with a retail planning solution which builds demand planning module in it.
They use 14 statistical time series algorithm to forecast sales. They use backtracking and simulation, two methods, to improve model fit based on mean absolute percentage error (MAPE).
It is a tool Used by planners in retail company to forecast their product’s sales which earlier they were doing using excel spreadsheet.
This tool uses 14 time series algorithm for all the products and the algorithm with least Mean absolute percentage error wins.
MAPE is calculated as (((Actual Sales - Forecasted sales)/Actual sales) *100).
and 1-MAPE = Accuracy (or model fit)
What backtrack and Simulation does is:
Backtrack (backcast) is the first step taken once the preliminary definitions have been established. Backtrack is a process by which the engine applies its forecasting method to a point earlier in the series. Backtrack will be optimistic (best case) of the forecast. Backtrack will look to the historical series, apply a forecast and then comparisons can be made to the actual data points further along in the series. Backtrack will self-correct as to better align with the known historical data points.
Simulation will run concurrently in the order of operations with backtrack and is effected by the same prerequisites in the order of operations. Simulation is very similar to backtrack with the discernible difference that simulation will not self-correct. Any errors with the forecast will therefore compound. Simulation can be described as pessimistic (worst case) of the forecast.
And finally a ration of 50:50 is taken for both Backtrack and simulation.
Kindly let me know if my sentences are not clear
Hi @NISHIKET, that’s a good explanation. I’m sorry but I may not be the right person to answer your question . Maybe someone else could help you.
thats no problem … thank you so much