Time Series Forecasting

[Hammad_Rashid]

Hello everyone!

I am working on a project, where I need to show how Alteryx can help us in sales forecasting.

Please see the attached dataset. It has weekly sales for 3 stores for the period Jan22 to Oct24.

Let's take historical period as Jan22 to Sep23. So, future periods will be Oct23 to Oct24.

Could you please create a workflow using ARIMA / ETS tools and forecast the weekly sales for the future periods?

Finally, let's compare how good is the forecast, when we compare it with original sales, we already have for Oct23 to Oct24.

Next, could you please create a macro for this, as in reality I will be forecasting the sales for 300+ stores. Also, please structure it in a way that I will have these 3 columns as the input, and whatever new store I want, I can stack it below and run the macro, so it will give forecasts for all those stores.

Thank you!

Hammad

[alexnajm]

Here is an article to help you in your journey: Performing Time Series Forecasting in Alteryx Designer. The video being referenced int he article is here: Performing Time Series Forecasting in Alteryx Designer

Here are some interactive lessons that could help too: Interactive Lessons - Alteryx Community. Here's another article as well for the batching portion that you are requesting Time Series Factory Tools vs Batch Macro Time Series

This is quite a large project to build a full workflow - please use the resources from the Community to build yourself and if you have questions along the way, then we are more than happy to answer. Good luck!

[Hammad_Rashid]

Hi Alex,

I have started building the workflow.

I did that for one store. 

Started with comparing ARIMA vs ETS and picked ARIMA.

Then, forecasted the sales data.

You will see it gave very good results for the 1st forecast quarter (Q4.23), but it just got flat after that starting from (Q1.24).

What could be the reason?

[alexnajm]

You don’t have enough past data to forecast that many periods out - you are using only 80 weeks or so of data to forecast 52 weeks. More historical data and/or less of a forecast would help