TS Forecast Model

[M-Harbi]

Hello

I am trying to build a forecast model that can predict future hardware demands for each business organization and each asset type combination.

I have three years data and the number of demands per org, per asset, per year, and I am trying to build a forecast model based on that info.

The results I am getting assume that each of those combination is a different year, which is wrong, I only have three years and many org/asset combination and I am trying to make the model to build a standalone prediction for each of those combos I have.

I am using TS Model Factory. Below is a small sample of the combos.

Any idea if I am using the right tool or how to build the iteration for each combo?

Org ID	Asset ID	Year	Sum_Transaction
1	5	2022	1460
1	5	2023	645
1	11	2023	282
1	11	2024	250
2	10	2024	1
2	11	2022	17
2	11	2023	93
2	11	2024	39
7	5	2022	2904
7	5	2023	3783

[OTrieger]

@M-Harbi 
Thanks for sharing your query.
Please share the WF that you developed so far so we can have an idea where you are facing the issue with this model.

[M-Harbi]

Thank you for your reply.

Please find it attached, I summarized it and removed the cleansing steps.

[alexnajm]

For the TS Factory tool to work, you need a group (org & asset), a Date field (Year), and a target metric field (Sum_Transaction) - it looks like you only have 1 of the three configured properly.

1. Because the TS Factory tool only accepts one group, concatenate them together with a Formula
2. Convert the Year into a Date field, i.e. 2023 --> 2023-01-01
3. Target field is good so far 👍

Make these changes and let us know if that helps - the only other issues I can see are a) you don't have enough years of data so your forecasts won't look good and b) you need each group to have a full amount of dates (i.e. can't be missing 2023)

[M-Harbi]

Hi Alex,

I tried your steps and it seems I am getting some results per combo, just one thing I wanted to verify, TS Factory Model only provides one year forecast? it cannot be iterated to provide 3 years forecast based on this data and the provided group?

Much appreciate your great help.

[alexnajm]

It depends on how many periods you select in the TS Factory Forecast tool! You should be able to do more periods, although the amount of data you have for each group may not make the forecast useable

[M-Harbi]

Really appreciate your help Alex!

The data I am providing is just a summary of the year, however the actual data contain thousands of records per requests, we will apply it once we have the right tools and configuration applied.

I attached a new workflow after the suggested modification, and I need some help with the following two points:

1- The TS factory model is showing the group intercept once, not sure if the three years are configured.

2- is the TS Forecast step needed here, since this is the only node that gives me three years results? If yes, it gives me 1 intercept per each year while we need each group intercept per year.

I really do appreciate your help and support. 

[alexnajm]

I think the issue is you are missing data for certain groups - i.e. 1-5 is missing 2022 data. This is not good and needs to be filled with something as each group needs to have the same amount of periods

You need the TS Forecast Factory tool, not the TS Forecast tool

[M-Harbi]

Ok I added some dummy data to fill up the missing groups per year, and inject it into the model.

Still I am getting one year intercept, and not getting the sup-periods to forecast three years, I attached the workflow.

Any thoughts?

[alexnajm · Accepted solution]

I'm getting errors too where it's not producing a model per group - I still think your data is not big enough to do an actual forecast. 3 years of data per group can't produce an accurate forecast

If you have it at a lower level of period (i.e. month), it may be worth forecasting at a monthly level and rolling up afterwards to the year