0
$\begingroup$

I am bit confused regarding the application of 10 fold cross validation steps. To be specific, I have made a multiple regression model (except model validation) and the model does not predict reasonable. So I think I have to do model validation for the prediction. I know that 10 fold cv method will split the data into 90% (training) and 10% (test). After that it will take 10 different attempts and then I need to take the average of that MSE of the 10 folds.

I want a step by step procedure for the application of Multiple regression for prediction or you can suggest me any link for that.

Improve this question
$\endgroup$

1 Answer 1

Reset to default
2
$\begingroup$

Cross-validation is primarily a method to tell you how bad a model is likely to perform in new data from the same "stream". Note that the optimism bootstrap is better because it validates the full sample fit and not a 0.9 sample fit as with 10-fold CV. To be as accurate as the bootstrap you need to repeat 10-fold CV 100 times and average. CV is not intended for developing another final model but for estimating the likely future performance of your original model (or at least for a 0.9 approximation of it).

Improve this answer
$\endgroup$
5
  • $\begingroup$ I am sorry but I am a non statistician. If you could provide me any link for multiple regression which encompasses the full details from data collection to model validation...I need this model for prediction. $\endgroup$
    – Arvinder
    Nov 27, 2014 at 13:33
  • $\begingroup$ There are many excellent textbooks and lots of courses at lots of universities. $\endgroup$
    – Frank Harrell
    Nov 27, 2014 at 13:37
  • $\begingroup$ Thanks for getting back to me. I searched exhaustively in google for the explanation but I was unable to find the full details. Is there any link which can sail me through all the details? At present the problem I am facing is I have made a multiple regression model without doing model validation. I was able to get the equation after satisfying the regression assumptions but I am unable to predict taking the confidence interval. So how should I go from here? Moreover the number of data points for the created regression model is 90. $\endgroup$
    – Arvinder
    Nov 27, 2014 at 13:40
  • $\begingroup$ what is the need of cross validation or bootstrapping. We can straightway take any out of sample data for prediction. If the response lies in the confidence interval then it means that we have built a good model otherwise not. Isn't this correct? Please advise $\endgroup$
    – Arvinder
    Nov 27, 2014 at 14:12
  • 2
    $\begingroup$ It is unreasonable to expect to be able to do a complicated task well without being willing to invest in studying the methods. $\endgroup$
    – Frank Harrell
    Nov 27, 2014 at 17:21

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.