6
$\begingroup$

Both can be used to select a best model and both penalize overfitting. However, CV is easier to understand and to apply in practice. Thus, when should I use AIC instead?

$\endgroup$
2
  • 5
    $\begingroup$ Isn't that a duplicate? I think this question might have been answered before. If not, then here is a brief answer: AIC is faster and often it is easier to apply in practice because AIC is part of the standard model diagnostics. There is no need to run CV if you already have the AICs of the competing models. However, AIC may behave poorly (the AIC values may be little relevant) if the assumed error distribution is far from the realized residual distribution. CV does not depend on that. $\endgroup$ Commented Mar 14, 2016 at 10:52
  • $\begingroup$ check this thread - stats.stackexchange.com/questions/48714/… ... it might answer your questions $\endgroup$
    – akashrajkn
    Commented Mar 14, 2016 at 11:29

0

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.