1

It is i'm sure quite a simple question but i was not really able to find an explanation of this online and does not seem to be explicitly called out in the typical expositions i've seen.

I get how 10 fold cross validation works to recursively partition the dataset into 9 train and 1 test set 10 times to get an estimate of out of sample performance.

What i'm not sure is how the 10 models are then aggregated into one single model that can be used to predict against a new dataset.

e.g. say i do 10 fold linear regression. my understanding is that i then have 10 individual models (with individual coefficients, se, p-values etc.) built on 10 subsamples of 9/10's of the data - is the 'final model' just an average of each of the individual models e.g. just average to coefficients of each model? This is the bit i was never quite sure on...

andrewm4894
  • 223
  • 1
  • 7
  • This, and related, questions were discussed on CV several times, please see [one](http://stats.stackexchange.com/questions/11602/training-with-the-full-dataset-after-cross-validation), [two](http://stats.stackexchange.com/questions/2306/feature-selection-for-final-model-when-performing-cross-validation-in-machine), and [three](http://stats.stackexchange.com/questions/27750/feature-selection-and-cross-validation) -- all three are on the first page of most voted questions with "cross-validation" tag. – amoeba Feb 25 '14 at 23:34
  • Thanks for the links - tried searching via a query but did not realize looking for top voted questions would also be a good way to search. Please feel free to close as duplicate of the suggested question. – andrewm4894 Feb 26 '14 at 07:51

0 Answers0