Predictive analytics, ARIMA model with forecasting for many programs

[Kmassey]

I'd like to use predictive modeling to identify which programs may lose money in the future. I have monthly data within the FY for hundreds of programs, in categories of actual revenue, actual margin, forecast revenue, forecast margin, etc. Each category is recorded per month, so for actual margin, I have values for January through October, actual revenue Jan - Oct., etc.

What I'd like to do is set up a process by which I run the analysis for all programs, then use the resulting forecast file to see which programs are forecasted to lose money.  In the process attached, I filter for Actual Margin, fill any empty months, run the data through the ARIMA tool and then the forecast tool.  This gives me usable data when my initial filter also filters by a single program, but meaningless data without the program filter.

Is there a way to run the forecast process iteratively to get results per program?  Or, is there a whole better way to approach this?

Thank you!

[christine_assaad]

Hi @Kmassey 

You can try the TS Model Factory tool available on the public gallery that allows grouping.

https://gallery.alteryx.com/#!app/TS-Model-Factory/5772af65aa690a1348cc6abf

This is how its config window looks like. Hope this helps. Cheers!

[Kmassey]

Wow, looks promising, but totally different from what I've used in Alteryx so far.  Is there a knowledge base you can point me to to interpret this macro? I've not used macros before.

[christine_assaad]

Hi @Kmassey 

Once you download it, it will appear in the  "Time Series" suite next to the other Time Series building blocks you are currently using. See below:

You will simply drag it, drop it into the canvas and configure it the same way you would do with other building blocks.

This is also the link to the Help page that talks more about this tool https://help.alteryx.com/2020.2/TS_Model_Factory.htm

Cheers!

[Kmassey]

Thanks, it took a bit of gyrating to get the tool loaded and available for use in Designer, something to do with location of the downloaded file in the Alteryx macros folder.

When I run the tool, I get the "insufficient records to run an ARIMA model" message.  My date field is correctly formatted, but my data are typically good for only one FY.  If I duplicate the FY data so that I have 2 year's worth, the model runs without error, but results and forecast are suspect, right?  For my need to forecast values based on 12 months of data, is there a better approach or tool?

[RolandSchubert · Accepted solution]

Hi @Kmassey,

ARIMA combines autoregressive processes with the moving average, for the approach the number of available periods is really very small, especially when it comes to detecting a correlation with values from previous periods. I think increasing the number of periods by copying them does not lead to useful results here. Maybe it would be better to simply work with a moving average? What do you think?

Best,

Roland