0
$\begingroup$

A model is trained to predict the median temperature of Boston. The resulting model works well according to their validation data. However, this model performs poorly when used to predict the temperature of Washington. Explain the reason and suggest a way of training a better model for Washington data.

I think these two datasets are not identically distributed, so the model obtained on one dataset is not generalizable to another dataset. The solution is that we should merge these two datasets and then do cross validation and train the model. Is it true?

$\endgroup$

1 Answer 1

2
$\begingroup$

This should be due to the fact that ML Models fails to give optimal results when distribution of data changes i.e. Data is not identically distributed. To solve this problem yes the best approach would be to merge the data and create a model on data from both states. This will make sure that your training distribution aligns with test/ real world data distribution hence improving model performance

$\endgroup$
1
  • 3
    $\begingroup$ Also try training separate models. $\endgroup$ Commented Oct 20, 2022 at 6:54

Your Answer

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

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