0
$\begingroup$

Briefly, I am training a model using XGBoost to predict future quantity for the factory to produce. Basic features currently in use are date time features, categories, holiday (binary). I have just done a prediction for the next period (12 months in year 2024). However, after presenting these figures to my supervisor, he said the prediction is far too low given the fact that the factory has just been upgraded with loads of new machines and thus, the capacity is surely increased in 2024.

Two different solutions come up to my mind:

  1. Incorporate a new feature reflecting the development of machines in the factory (if not possible, use the potential development of the factory's capacity instead or anything else that illustrate the development of business) then backfill other data points in the past
  2. Post-process the prediction of the current model, i.e, if the predicted quantity of January, February, March 2024 is 5, 10, 15, then they are multiplied with a constant, e.g 3, then become 15, 30, 45 consecutively (me and my supervisor will discuss about an appropriate constant)

What is the right way to do in this case?

$\endgroup$
2
  • $\begingroup$ You seem to focus in a very specific use-case. Please introduce that use-case in the beginning. Furthermore, it would be interesting what model you use and what type of training data you have. $\endgroup$
    – Broele
    Commented Sep 26, 2023 at 16:54
  • $\begingroup$ Thanks for your comment, I will update the information below in my original post.In my case, I am training a model using XGBoost to predict future quantity for the factory to produce. Basic features currently in use are: date time features, categories, holiday (binary). $\endgroup$
    – Bourbon
    Commented Sep 27, 2023 at 0:20

0

Your Answer

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