This famous dataset have a lot of zero values (Above 500). But it is not really clear, some of them are outliers or not.

My question is: What if i decided that ALL 0's are outliers (it's pretty likely) and removed objects with any 0 from the dataset and faced with zeros in unseen test dataset (in kaggle competition i.e). I can't just delete an object from the test sample. Is it even possible or my assumption was initially incorrect?

I am interested in it, because of this solution. Some users doubted one of the steps (imputing mean instead of zero value according to the target class). It there are some zeros in test data, than the model isn't correct, else there's nothing to stop exposing the information to the distribution



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.