Best way to know correct forecast model - process chain set up with multiple forecast models

[sourabh_jain9]

Hi Experts,

I need your help in selecting best forecast model for our company. We have some of the models already used for our company, and because of multiple models used it is taking very long time for process chain to finish. There is no existing documentation available on which model was used why initially. Please help me to make out forecasting process smooth.

- What is the best way to know, which forecast model is correct and should be used for our forecasting process.

- In case multiple forecasting models are really required to be used, please suggest ways to optimally schedule them in process chain.

- At times we get messages like "not enough data available" for specific model - any way to avoid this.

- How to optimally use parallel processing profiles forecasting process in process chain.

- Things which should be avoided.

Request your help, please share your experiences.

Regards

NB

[Former Member]

Hi NB,

Depending on ur company's requirement you have to select the models so as to reduce complexity.

Let me know what level of complexity are you looking for your models, they it can help me in helping you.

Thanks

Ankit

[sourabh_jain9]

Hi Ankit,

Thanks for your reply here, i am looking for moderate complexity for forecast model.

Anything which gives me as accurate a result as possible, the thing is we can't afford too much of a run time for model to run with process chain. We are already using parallel processing for process chain run.

Few other point, if you can help me on those as well.

- We are already running process chain and statistical forecasting with multiple models. Currently the run is daily, please let me know if the daily run is really suggested. 

- At times we get messages like "not enough data available" for specific model - any way to avoid this.

- How to optimally use parallel processing profiles forecasting process in process chain.

- Things which should be avoided while selecting forecast model.

Rgds

NB.

[rahul_agarwal5]

Hi Neelesh,

There are many points you need to consider to redesign forecast process for your company/client.

You need to select the best suited forecast model first depending on the business. This has to be well tested & agreed by business users. Complexity will be an outcome of this exercise with business users. Best id to give then a brief intro on all available models & then help them selection the best one as per their requirement.

Auto selection models are generally more time taking & should be used only when you have no idea at all on the business/demand pattern.

Run time will depend how you are clubbing the CVCs to get the forecast generated & also parallel processing. For parallel processing profile you will need to do trial & error testing along with help from Basis team on how many free dial up processes are available.

Even you can run many forecast calculations in parallel if the product/cvcs are totally different. - As per my personal experience maximum run time reduction can be achieved here.

Daily run is not advisable except only for businesses where you have too much dynamism in demand planning i.e. you expect the demands to be changed overnight. Most of the companies run forecast on monthly basis or at weekly basis at the max.

"Not Enough data" will be a problem if you are having the irrelevant models used in forecast profiles. This means users are not bothered to maintain the needed data for he forecast calculations or they are not aware at all of the situation. Running such models on daily basis is not advised at all. Better users should use interactive forecasting & saving the results in such cases.

Just to give a crude example we get forecast calculated on monthly basis for approximately 4 lac cvcs in less than 3 hrs using moving avg, seasonal linear regression, seasonal trend, croston models. We use parallel profiles also everywhere with 10 blocks & 500 cvc/block.

Hope this helps. Let me know if you have nay more questions & also the results using any of this.

Regards,

Rahul

[sourabh_jain9]

Hi Rahul,

Thanks a lot for detailed reply, i will check as per your suggestions in the system and get back to you for any further query.

Rgds

NB

[sourabh_jain9]

Hi Experts,

Is there nobody who can help me in this. I really need your help, please reply.

Rgds

NB

[Former Member]

Hi,

Instead of using multiple forecast model, can you please try running automatic model selection and compare the result.

check : Demand Planning and Forecasting - Forecasting & Replenishment - SAP Library

In case of running multiple model, you can use different keyfigure for each model, create activity to run each model and schedule job in background.

You can use parallel processing profile.

You should not run multiple model in parallel for same selection.

Regards

Kishor

[sourabh_jain9]

Hi Kishor,

Really appreciate your help here, i have actually gone through multiple documents on this and configuration settings are clear to me.

What i would like to know is how to know which statistical forecast model is suitable for our company.

Any help here will be really appreciated.

Rgds

NB

[sourabh_jain9]

Hi Experts,

Can anybody please help me, really appreciate your help here. Please share your experiences with me.

Rgds

NB