Can bootstrap is used for estimation prediction error



I am currently studying about bootstrap and while studying it I came across question can bootstrap is used for estimating the prediction error.I have searched it and found it that

To estimate prediction error using the bootstrap, we could think about using each bootstrap data set as our training sample, and the original sample as our validation sample.

What I want to know is it a correct way and to use bootstrap to estimate the prediction error and what are the problems associated with that.


Hi @ankit81195

Where did you find this?? That is rather surprising as 66% of your original observation will be in each bootstrap you it means you will do a test on minimum 66% of the data you used to do your model !!! You will have great results for sure but the error on the validation set will be something else.

I hope you did not find this on a textbook let me know which book



This question appears in this Stanford document

I found it by googling it. The same document contains the answer.

Hi Ankit, if you are reading this document or some copy of the document, please read the full section. If you find the answer, please post it here for the benefit of other readers.


hi @r_achar

got it now, the paper refers to cross validation, I was referring in case you train your model with bootstraped data sets and each training set is based on you whole initial sample. In the case of cross validation we do a bootstrap on the training set only, which mean that you test set is never used during the training stage,

Thanks to share this paper really good paper.

Have a good day.