2
$\begingroup$

I have implemented a classifier and I can calculate the precision and recall when testing the classifier (two classes) in the test set(>100). Is it meaningful to talk about the confidence interval for the precision(recall)? The precision(recall) can be defined as a random variable. I assume that my sample is the test set and there is a precision related to it with the defined classifier. How should I define the distribution for the precision for class 1 (#of correct classifications of c1 / #of total classifications of c1)? One alternative is to use the Binomial distribution for proportions.

Improve this question
$\endgroup$
2
  • 2
    $\begingroup$ I would opt for the bootstrap to compute such confidence intervals. $\endgroup$
    – Marc Claesen
    Oct 5, 2014 at 10:22
  • $\begingroup$ Yes, certainly you can talk about the confidence interval for precision or recall. $\endgroup$
    – user31264
    Oct 5, 2014 at 10:47

0

Reset to default

Your Answer

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