0
$\begingroup$

Goal - Predict number of days the finished good would be delayed from a promised date of delivery?

Background - It is only 7 weeks before the promised date of delivery that the demand becomes proxy final. Meaning there can be a change in Qty/promised date of delivery/both in the 7 weeks period (run up to promise date of delivery)

Challenge- If my prediction point is (Promised Date of Delivery - 7 weeks). It is likely that some of the feature value might change in the 7 weeks period.

Question - What approach or how do I cater if the following happens after my initial prediction?

  1. Qty changes (new qty < initial qty), after my prediction? - Scenario -I

  2. Promise date of delivery is moved further? Scenario -II

  3. Both the qty changes and promise date of delivery is moved further? Scenario -III

If I have a secondary model for the following scenarios :

a. How would I cater for the 3rd change, will it be catered by my second model?

b. What will be the composition of the training set? Training Set Model I - Initial state of the demand Training Set Model II - Second State of demand +3rd State of demand.. + Nth State of Demand

or

Do I need a different approach to tackle this problem and what would it be?

enter image description here

enter image description here

enter image description here

$\endgroup$
7
  • 1
    $\begingroup$ You should just do a new prediction with this new/updated data. Of course it depends if you have included this features in the data set yet ? You could include features like nb of changes already done yet ... But please give details about your features and model. $\endgroup$
    – Malo
    Sep 11, 2021 at 9:30
  • $\begingroup$ @Malo - I have added more information to the question, hope this gives more context. I have thought of adding the feature, but not sure if it would be sufficient based on the dynamic nature of the process. $\endgroup$
    – Jas999
    Sep 11, 2021 at 10:57
  • $\begingroup$ @Jon Nordby I had gone through your answer in link . The problem is on the similar lines. I was wondering if you had any advise on the best way to tackle this? Thanks $\endgroup$
    – Jas999
    Sep 11, 2021 at 11:53
  • $\begingroup$ You shoud focus on: * only 1 model: it seems you do not need to have more if proper features are used * more features like: initial quantity, replanned quatity, delta qualtity, initial delay, replanned delay, delay delta, nb of replanning, company/team workload, holidays periods, nb of employees ... * gather data about past real cases $\endgroup$
    – Malo
    Sep 11, 2021 at 13:11
  • $\begingroup$ @Malo Thanks. But what would be my point of prediction in the above process? If I make the prediction seven weeks prior to date, the variables such as replanned quantity, replanned delay would be zero. Are you suggesting to recalculate if there is a change in the feature value and have a fresh prediction? $\endgroup$
    – Jas999
    Sep 11, 2021 at 19:45

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.