Talk:Precision and recall

The article lead immediately refers to "instances" without providing any context or explanation of what "instances" are being referenced here.

dr.ef.tymac (talk) 00:56, 6 August 2021 (UTC)Reply

Also, the formula for precision and recall in the intro are the same when they should not be. Both are defined as "relevant retrieved instances"/"all relevant instances". 2603:6011:3E0B:55EF:C0E3:6E88:CB56:8439 (talk) 01:52, 2 March 2024 (UTC)Reply

Never mind, I miss-read it 2603:6011:3E0B:55EF:C0E3:6E88:CB56:8439 (talk) 01:53, 2 March 2024 (UTC)Reply

In the Astronomy community, precision and recall are often called completeness and purity (examples: Lochner et al. (2016) - page 8, Boone (2019)). I would like to include them with the other synonyms in the table. Does anyone know to to include them? Cyber Mulher (talk) 01:50, 5 February 2021 (UTC)Reply

In my opinion, the current use of the graphical illustration is not optimal. The German version of this article [1] uses the same picture, but additionally has two variants of the image (see [2]) which illustrate the individual concepts (P & R). The German article also uses the same colors in the True/false positive/negative table as in the image, further illustrating the connections. We should make those changes once we've agreed on how to do the merger (see below). Tobi Kellner (talk) 19:34, 1 July 2008 (UTC)Reply

I still don't understand the illustration. Perhaps a more thorough explanation of the color and arrows would help. —Preceding unsigned comment added by Toahi (talk • contribs) 00:43, 27 February 2010 (UTC)Reply

Agreed. I added descriptions of the regions to the caption a few days ago. --Vaughan Pratt (talk) 19:46, 7 August 2011 (UTC)Reply

The illustration is incorrect, precision and recall should be switched. 150.145.127.42 (talk) 13:58, 10 June 2024 (UTC)Reply

It was my mistake, I was interpreting it incorrectly. The illustration is correct 150.145.127.42 (talk) 14:11, 10 June 2024 (UTC)Reply

I am suggesting that the Precision (information retrieval) and Recall (information retrieval) be merged into this article. A similar movement between sensitivity and specificity is being discussed at Talk:Sensitivity_and_specificity#Merger_proposal, and it seems like the consensus is heading toward a merger. WDavis1911 (talk) 18:26, 6 June 2008 (UTC)Reply

I absolutely agree. We just went through a similar debate at Talk:Relevance (information retrieval) and I'm happy with how we redirect discussion of performance measures to Information retrieval#Performance_measures, which in turn points to this entry. Dtunkelang (talk) 04:53, 10 June 2008 (UTC)Reply

I completely agree. I started this article in September 2007, probably because I found that there was no English article that matched the German one on precision and recall [3]. Now I realize that we have a lot of redundant information, with the separate articles on Precision (information retrieval) and Recall (information retrieval) as well as the section Information retrieval#Performance_measures. I think there is a point to be made for having a discussion of precision and recall in one place, rather than simply having two separate articles, though, as the two terms are so closely related that it seems to make sense to explain them together. But maybe this article should have clear separate sections with effectively the contents of Precision (information retrieval) and Recall (information retrieval) for those looking for just a definition of one of the terms alone. Tobi Kellner (talk) 19:34, 1 July 2008 (UTC)Reply

I have had a go at doing the merger, basically just copied the text from the other pages. But I am a bit inexperienced so not sure what to do with the other pages Precision (information retrieval) and Recall (information retrieval) OZJ (talk) 16:27, 24 June 2009 (UTC)Reply

Mind the fact that there is now a Evaluation measures (information retrieval) page. This requires a bit of coordination. i⋅am⋅amz3 (talk) 00:49, 18 March 2018 (UTC)Reply

Would it be appropriate to say something about the relationship between precision/recall and soundness/completeness? 129.240.71.209 (talk) 12:35, 18 May 2010 (UTC)Reply

This page is very confusing. Here is a link to a good explanation:

http://newadonis.creighton.edu/hsl/searching/Recall-Precision.html —Preceding unsigned comment added by 171.66.73.218 (talk) 00:10, 6 April 2011 (UTC)Reply

I think the first sentence "Precision and recall are two widely used statistical classifications." is imprecise and potentially confusing given the relation of these terms to statistical classification. More accurately they are metrics of performance for statistical classifiers, not really "statistical classifications". --Jludwig (talk) 06:23, 10 June 2010 (UTC)Reply

I noticed that Fall-out in the table of metrics for definition points to a wiki on information retrieval rather than a definition for fallout or a class of metrics covering it. — Preceding unsigned comment added by 2620:0:1009:18:DCAF:5105:B547:5A5E (talk) 19:16, 22 August 2018 (UTC)Reply

The equations don't include cases when the denominator is 0. Are precision and recall just undefined in these cases? Khatchad (talk) 23:09, 7 March 2011 (UTC)Reply

Technically yes. But in that case the numerator is necessarily zero so the result would always be not-a-number (NaN) rather than infinity. A harsh grader is someone who avoids NaN by taking precision and recall to both be zero when nothing relevant is retrieved (which covers all 0/0 cases), a kind grader would take precision to be 1 when all retrieved items are relevant and recall to be 1 when all relevant items are retrieved and otherwise does the same as the harsh grader (hence also covering all 0/0 cases). --Vaughan Pratt (talk) 00:11, 4 August 2011 (UTC)Reply

The article on accuracy and precision talks about pretty much the same concepts as this article, but does so differently, and completely fails to mention precision. This article covers all the concepts, but doesn't have the nice diagrams (e.g. the bullseye) of the former. But essentially, they're pretty much about the same thing. Thus, a proposal to merge these two articles. Good idea, yes or no?

Failing the merge proposal, the whole, ahem, mess of related articles could benefit from more cross-links and sharing: e.g. accuracy and precision fails to link to this one when defining recall, and instead links to sensitivity (tests) for recall. And likewise, as one chases around the various links in this (non-)cluster of articles. Arghh, so e.g. at the bottom of sensitivity and specificity is a table, defining precision, recall, accuracy, and many others, but completely forgetting to mention F1! The article on information retrieval points to this one as the "main article" on precision, recall, and fall-out, but this article never mentions fall-out. So maybe not just a merge, but a coherent rationalization of the whole cluster of related topics? linas (talk) 19:21, 2 June 2012 (UTC)Reply

Oppose merge I disagree wholeheartedly with the merger idea; physicists and engineers talk about precision and accuracy, computer scientists talk about precision and recall. It makes sense to explain the concepts behind these terms in terms that the respective separate communities are familiar with. Or in other words, if I want to find out the difference between precision and accuracy in one context (physics), having to learn one more term (recall) in a different context (machine learning) will not make it easier to understand the concepts as they apply to physics. But perhaps old discussion = dead anyway, and the merger proposal should be considered deceased by lack of attention? Anonymous Coward 21:40, 6 June 2012 (UTC) — Preceding unsigned comment added by 77.87.228.68 (talk)

Comment I am for the rationalization of the cluster, and I am pretty sure that the valid concerns of the above unsigned comment can be addressed with appropriate craftsmanship. But as always this is a fair amount of work; the cluster needs to be read carefully, and the contexts in which other articles link into the cluster should be examined. If I'm not mistaken, the cluster we're discussing rationalizing consists of this article, Accuracy and precision, and Sensitivity and specificity, right? My snap judgement is that this article, P&R, is very similar to S&S, but not nearly as similar to A&P. P&R and S&S are about classification or decision procedures, and A&P is about quantitative measurement. Thoughts? ACW (talk) 18:21, 25 June 2012 (UTC)Reply

Oppose merge I just want to add my opinion that I agree the Accuracy and precision article is about quantitative measurement, not information retrieval, and so definitely should not be merged with this one. The defining characteristic of precision in the other article is reproducibility, while here it is relevance. Two quite distinct concepts. ArthurPSmith (talk) 17:10, 29 June 2012 (UTC)Reply

Oppose merge I want to forcefully add my voice to the dissenting opinion. Not wanting to offend anyone, but the desire to merge these topics reflects a limited appreciation of their relevance to other disciplines. Perhaps there should be more cross-links among the articles to benefit the Information Retrieval community (and others), but Precision and Accuracy are important, long-standing concepts in quantitative measurement and deserve an entry independent of Information Retrieval. I'll take your word for it that various Information Retrieval pages could be improved, but please don't do so by gumming up Precision v. Accuracy.129.82.37.172 (talk) 19:10, 9 July 2012 (UTC)Reply

Oppose merge The main argument for this merger is that the articles are confusing. This is not really the case. The article on accuracy and precision is very clear and succinct. It is in fact so much so that this is one of the only pages that I would feel comfortable citing when looking for definitions and differences between the two terms. — Preceding unsigned comment added by 130.182.25.202 (talk) 00:46, 6 July 2012 (UTC)Reply

Oppose merge Do not merge. They are different articles. Pkgx (talk) 17:06, 12 October 2012 (UTC)Reply

Oppose merge This is a very bad idea. The articles use the ambiguous term perception in two completely different senses, as the first anonymous user above has pointed out. Accuracy and precision are statistical measurement concepts used in all physical sciences, as well as psychology. This article seems to be about concepts exclusive to psychology. I think enough time has passed with no consensus to merge, to remove the tags from both articles. JustinTime55 (talk) 19:14, 24 October 2012 (UTC)Reply

The table displayed is incorrectly labelled. — Preceding unsigned comment added by 128.192.34.23 (talk) 17:45, 24 July 2014 (UTC)Reply

I would say the figure is incorectly labelled, false positives should be false negatives — Preceding unsigned comment added by 81.82.250.24 (talk) 10:50, 7 December 2016 (UTC)Reply

The relationship between sensitivity and specificity to precision depends on the proportion of positive cases in the population, also known as prevalence; with fixed sensitivity and specificity, precision rises with increasing prevalence.

Seems to be wrong, if the prevalence is increasing the precision doesn't necessarily increase or decrease since the precision is the number of true positives over the total number of positives. Let me know if I misunderstood. PeepleLikeYou (talk) 11:28, 15 August 2020 (UTC)Reply

The quote is correct. The precision depends on the prevalence. One way to see this is to imagine what happens if the prevalence goes to 0 while the sensitivity and specificity remain constant. Since the precision is the number of true positives over the total number of positives, as you said, it must fall to 0 too (there can be no true positives if the prevalence is 0). Tobycrisford (talk) 11:25, 7 January 2022 (UTC)Reply

The sensitivity is equal to the number of true positives over the number of relevant elements. It is not possible for the sensitivity to remain constant as the prevalence goes to zero. PeepleLikeYou (talk) 00:58, 8 January 2022 (UTC)Reply

Fellow Wikipedians: I've proposed some changes to the formula infobox transcluded into this article, with the goal of trimming down its overpowering (if not excessive) width. My original message with some explanatory notes is at Template talk:Confusion matrix terms#Template_width, and you can see the revised template layout I've proposed by viewing its sandbox version.

There have been no responses over there in well over two months, and since the changes I'm proposing are significant enough to possibly be contentious, I wanted to invite any interested Wikipedians to discuss them over at the template's talk page. Thanks! FeRDNYC (talk) 00:05, 5 January 2022 (UTC)Reply

https://maxkleiner1.medium.com/bayes-theorem-confusion-matrix-b6f9ee3864a0 Biggerj1 (talk) 14:52, 2 September 2022 (UTC)Reply

  This article was the subject of a Wiki Education Foundation-supported course assignment, between 20 January 2023 and 15 May 2023. Further details are available on the course page. Student editor(s): AbigailG23 (article contribs).

— Assignment last updated by AbigailG23 (talk) 05:13, 9 April 2023 (UTC)Reply

Many of the referenced articles appear several times (different reference numbers). It would be nice to consolidate those. 194.230.147.195 (talk) 20:17, 30 April 2023 (UTC)Reply

It seems odd to me that there is no precision-recall figure on this page, as this visualizes the trade-off of precision and recall. There is one on the F-score page. Ramajoepanda (talk) 14:08, 18 January 2024 (UTC)Reply