cancel
Showing results for 
Search instead for 
Did you mean: 

DP: Choosing a forecast model

Former Member
0 Kudos

Hi Gurus,

Iam working on a support project now..in the DP during the implementation time the forecasting functionality was not implemented(because the stat fcst is copied from an ext sys)..now the business wants to use the APO forecast functionality..

Can someone share your ideas(with an example) on how to choose the fittest/correct forecast models..we have over 1500 products now for which we want to use the APO functionality..

Thanks

Dhanush

Accepted Solutions (1)

Accepted Solutions (1)

Former Member
0 Kudos

Hello Dhanush:

Shortest Answer: Best Models are complicated=> Can you get the grouping of the

models that were imported into DP? Many Forecast Products may have made each of

your 15000 products unique models!  Someone in the business should give you some

help - or do they just want you to Best Fit DP and see what it can do?

Anyway...Here goes:

This takes some effort and there can be a significant effort to this. To do this well may

require someone with a sound statistical knowledge. This is For Batch Processing!

If you have Interactive Forecaster, they could be creating GUID Forecast Profiles

Dyananically attached to you Data Selections!!  (and could override the Forecasting Batch Process - this! This is a Configuration and Process Integration Question - as unexpected forecasts may results occurr!)   Interactive Forecasting could be used to create your own Master Forecast Profies/Models

for use in Batch: Be Careful => Users could override you MFPs/Models!!!

1. Product Grouping:  It would help to group similar Products/Product Groups with similar Demand History of thes 15000 Products.

If these Products are already in DataViews/Selections: That would be great - because you need

to assign/connect these to the Forecast Profiles

Forecasting is typically based on products with similar seasonality and trend. (Assuming no Customer as customers may add a degree of complexity)

2. Create Master Forecast Profiles

Each Group then may be assigned a Forecast Profile. Group and Assign a Master Forecast Profile/Model Univariate Profile (the Model Selection takes Place here * Use Best Model or

try to select a simple model

Group                            Count   

  I Consumer Products     #of Matls(CVCs)  Data_Selection=> Master Forecast Profile => ModUniv

         a Stable  Demand        2 ,000                CONSTBDMD       MFPCONSTB         CONUNI1                                                                

         b Seasonal Demand     8,000                 CONSESDMD      MFPCONSEAS       CONSUNIS2

         c New Products           2,000                 CONSNP    

II Commerical Products 

         a Stable Demand        1,000                  COMDMD       MFPCOMSTB         COMUNI1         

         b Seasonal Demand   2,000                COMSTBDMD     MFPCOMSEAS     COMUNUI2      

3. Create Master Forecast Profiles for Your Planning Area Using /SAPAPO/MC96B:

4. Create and Attach the Model Profiles to the Master :  This is where you need Forecasting

Expertise     

This really (for simplicity) involves picking a Univariate Profile and "adopting it" to the Master Forecast Profile.  I have kept these one to one ie One Univariate Profile  to One Master Forecast Profile:  Try Model 50 ie Automatic selection, 

A. Model Parameters (Try 50 and see if Automatic Selection Works)

    (I did not have much luck there...chose simple models to start)  

B.Diagnosis Groups (Alerts) may be set up to track Error in the 

Hint: Keep and establish a good naming convention                  

Note: For Model Selection: "Finding a Best Model"  Many do this outside of DP ie in Excel or some other Forecast Product and set the setting there

5. /SAPAPO/MSDP_FCST2: Attach Data Selections to Forecast Profiles

6. You can set up batch Jobs to run...

That is all!!! Best of Luck!

Steve

Steve Subichin

Former Member
0 Kudos

Thank you Steven ..

It is very helpful to me..

Best Regards,

Dhanush

Answers (0)