Talk:Accuracy paradox

Alternate titles
A random person writes about this article:

I have moved the discussion here from my talk page as this seems a better place for it. My impression is that the topic is known in different ways in different fields but it's not clear which is the most common. Andrew🐉(talk) 07:57, 18 March 2023 (UTC)


 * As far as I can tell from searches on google and google scholar the term "accuracy paradox" starts popping up around 2006. By now it is widely used especially in blog posts and tutorials on machine learning, but by and by also in scientific papers.
 * The term "false positive paradox" seems to be slightly, but not much, older. However it seems to be far less common by now because of the many mentions of "accuracy paradox" in the context of ML.
 * So if we keep the article the name should stay and "false positive paradox" be mentioned as alternative term in the article. Whether or not it is removed "accuracy paradox" should be added as alternative term for "false positive paradox" in the base rate fallacy article. In fact I just did that.
 * The term "accuracy paradox" is definitely common enough by now to warrant at least a redirect to Base rate fallacy. Most of the arguments from an earlier discussion that led to the deletion of an earlier article do not apply any more.
 * What strikes me most is that neither term seems to be used in statistics at all. It seems to me that "accuracy paradox" is a neologism that has come to be used widely in the context of machine learning. I get the impression that this is reflective of a lack in formal statistics training among many (aspiring) data scientists. E.g. of the three current references for the "accuracy paradox" article one is a self-published master's thesis in software engineering, one is a post from a data science blog, and one is a paper for a information retrieval conference.
 * That's why I think a redirect to Base rate fallacy is sufficient.
 * Random person no 362478479 (talk) 09:29, 18 March 2023 (UTC)