0
$\begingroup$

Im building two models, (model1 and model2) I trained and tested them on the same test dataset, model1 will have mean absolute error (MAE) 10.3, rooted mean squared error (rMSE) 30.1 model2 will have mean absolute error (MAE) 11.8, rooted mean squared error (rMSE) 29.5

Can any one help me to explain why one model is better in MAE while the other is better in rMSE? whats going on my data or models? Thanks!

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ They are different functions, and are not monotonic transforms of each other (unless your sample size is 1.) You should not expect them to preserve order across data sets or across models, although in my experience it is somewhat more common that they do than that they don't. $\endgroup$
    – jbowman
    Feb 4, 2017 at 1:43

1 Answer 1

Reset to default
0
$\begingroup$

If you trained your models using MSE then you should choose the one with the smallest MSE in your testing set.

Improve this answer
$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.