1
$\begingroup$

If I am doing a machine learning experiment A and my accuracy lies in the interval of 0.8 +/-0.03 and I have a machine learning experiment B with an accuracy in the interval of 0.9 +/-0.1, how can I combine these two results into one? Are there any standard methods?

(The accuracy numbers I gave are for test sets that contain a sample of a whole population. However, these are two totally independent populations and samples.)

$\endgroup$
4
  • $\begingroup$ just to be sure you are referring to accuracy with respect to a classification problem correct? $\endgroup$ Jan 8, 2016 at 20:01
  • $\begingroup$ Yes, these are classification experiments. Edited the question. $\endgroup$
    – You_got_it
    Jan 8, 2016 at 20:03
  • $\begingroup$ statistical ensemble? en.wikipedia.org/wiki/Ensemble_learning $\endgroup$
    – Mitch
    Jan 8, 2016 at 21:52
  • $\begingroup$ Thanks for the suggestion, @Mitch. What I am looking for is actually how to combine the results for reporting. So, I do not want to combine the classifiers or classification tasks, but rather, is there a way to calculate an overall accuracy for A and B? $\endgroup$
    – You_got_it
    Jan 8, 2016 at 21:56

1 Answer 1

2
$\begingroup$

The combine accuracy depends on the proportions of A and B experiments. If we call the proportions of A and B experiments $\tau_A$ and $\tau_B$,

the combine accuracy is: $ Acc = 0.8\times\tau_A + 0.9\times\tau_B $,

and the combine accuracy uncertainty is: $\delta Acc = \sqrt{0.03^2\times\tau_A^2 + 0.1^2\times\tau_B^2}$.

$\endgroup$
0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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