0
$\begingroup$

Hi I'm currently trying to predict if an item will be successful in my store, this means (How much is going to sale in USD) My training dataset contains many features:

  • Item name
  • Item weight
  • Item category
  • Item country of origin
  • Item sales overall
  • Item sales per store
  • Item rating
  • Item price

Etc.... Since I will be introducing a new item for sale, I know very little about this new item:

  • Item name
  • Item weight
  • Item category
  • Item country of origin
  • Item price

Not all the features present in training data/test data will be present when I will be making predictions. Is this normal in ML ? What is the rule of thumb when doing feature engineering for this type of cases.

$\endgroup$
1
$\begingroup$

Usually you can only predict with the variables you have trained on. But in a case like this, I would suggest you check the Multicollinearity of these missing variables with the ones you will have. May be they are already highly correlated with the features you already have. In that case, you can just model using the available features.

If that is not the case and the missing variable looks like a significant variable in your regression, then you might need to pick some other variables which can help you indirectly derive the missing features. For example may be category helps you derive the rating with some accuracy.

|improve this answer|||||
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.