Is there a fundamental difference from creating a model for each value in a category?

Is there a fundamental difference from creating a model for each value in a category?

Yes there is.

If a model is trained for each specific value of a variable (a category), then only the subset of data for this category can be used to train and test the model. As a consequence each model has a smaller number of instances to be trained from. Consequences:

• In the case of a small category, there might not be enough instances to obtain a reliable model.
• Every model is independent. This can be good or bad depending on whether this independence is also true in the data or not, or to what extent:
  • If the features behave in a completely different way depending on the category, then it's better to create individual models since each can really exploit the specific patterns for this category.
  • If the features have a very similar behavior across the categories, then independent models by category would potentially lose a lot of information.

In conclusion the choice often depends on:

• How much data is available for each category.
• How independent are the other features with respect to the category.