Does a model trained on a larger training set created by duplicating records perform better on the test data than the one trained on data without duplicates?


Duplicating training data for classes with fewer samples can actually improve your accuracy and that's what we call Oversampling.

But if you have balanced data and trying to duplicate then it doesn't differ from doing one more training iteration/epoch.

  • $\begingroup$ If you replicate your data by k folds, it is the same as doing k extra iterations/epochs on initial data. For balanced data sets, it may improve the accuracy but for unbalanced ones, experiments can only provide the answers. $\endgroup$ Oct 9 '20 at 7:31
  • $\begingroup$ I have tried a similar experiment for Regression with unbalanced data but the results didn't improve much. $\endgroup$ Oct 9 '20 at 7:39
  • $\begingroup$ Thank you, that was very helpful! $\endgroup$
    – Sid
    Oct 9 '20 at 21:44

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.