1
$\begingroup$

There are several concepts for classification performance : precision, negative predictive value, recall, and specificity.

Why do people often choose only precision and recall together ?

Why not negative predictive value or specificity?

Why do they choose F score in addition to precision and recall?

Can precision and recall alone deal with the imbalance problem between sample sizes of different classes?

Thanks!

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

Together, precision and recall represent a trade-off. You can increase one, but then you have to decrease the other. So, they are connected with each other, and together represent one dimension along which you can speak of optimization of your system. Drawing a curve to represent the relationship between these two can often be very revealing.

The use of an F score along with precision and recall helps to make comparisons between different situations. If you increase precision and decrease recall, is that a good thing, or not? Where do you stop? The F score gives you one tool to help you think about how to balance this dimension of the problem. So, it often gets used with precision and recall.

Improve this answer
$\endgroup$
6
  • $\begingroup$ THanks! Why not negative predictive value or specificity? $\endgroup$
    – Tim
    Commented Mar 11, 2014 at 3:14
  • $\begingroup$ Negative predictive value and specificity are just the flip side of precision (positive predictive value) and recall (sensitivity). I think that the main reason that precision and recall are used more often is that people tend to set up tests with a positive outcome in mind. This wikipedia link article has a good chart about half-way down that shows the relationship. $\endgroup$
    – John Yetter
    Commented Mar 11, 2014 at 3:26
  • $\begingroup$ but Negative predictive value and specificity can be determined from precision (positive predictive value) and recall (sensitivity), respectively. Do people not care about the negative outcome? Can it happen that recall and precision are good, while negative predictive value and specificity are bad? $\endgroup$
    – Tim
    Commented Mar 11, 2014 at 4:04
  • 1
    $\begingroup$ Yes, negative predictive value and specificity are definitely related to precision and recall, as you state. If you were to mess up precision and recall, you would also mess up negative predictive value and specificity. I believe that is why people get away with concentrating on only one side or the other. I think that people usually formulate their thoughts around the positive side, and so precision and recall are used more often. $\endgroup$
    – John Yetter
    Commented Mar 11, 2014 at 4:27
  • $\begingroup$ Sorry I should write "Negative predictive value and specificity cannot be determined from precision (positive predictive value) and recall (sensitivity), respectively" $\endgroup$
    – Tim
    Commented Mar 11, 2014 at 5:37

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.