topic Re: Detection des points atypiques/ Outlier detection in Discussions

Detection des points atypiques/ Outlier detection

VéroniqueB — Fri, 09 Jun 2023 00:28:15 GMT

Bonjour à tous,

Je souhaiterais savoir ce que vous utilisez comme méthode pour justifier qu'une valeur est atypique?

Moi j'utilise simplement la boxplot des outliers de Tukey, et en général je m'en sers pour justifier d'un rejet de la valeur . Mais est-ce que cela suffit?

Merci pour vos retours.

Re: Detection des points atypiques/ Outlier detection

Kevin_Anderson — Tue, 02 Feb 2021 00:20:08 GMT

Hi, VéroniqueB!

JMP has many opportunities to detect outliers. One step is under the Analyze pulldown > Screening > Explore Outliers. There are many different approaches to explore. There are also tons of reference literature on the subject, much of it written by titans. You can implement any procedure your heart desires with the JMP Scripting Language.

But the answer to the next question, what will you do with them once you've found them, is not so straightforward, and is very dependent on the context of what problems you're trying to solve. The rejection of outliers on a purely statistical basis is a dangerous procedure, and is not to be taken lightly. Many scientific discoveries (planets, chemical elements, etc.) have resulted from the investigations of outliers. Maybe the data homogeneity is the illusion, and the outliers have shone the light of truth on the problem with the data generating process. No available outlier rejection criteria are superior to the judgement of an experienced investigator who is thoroughly familiar with their measurement process.

Enough haranguing! Good luck! I hope you discover another planet or chemical element!! :)

Re: Detection des points atypiques/ Outlier detection

statman — Tue, 02 Feb 2021 02:28:29 GMT

Wow...Fantastic answer. Taguchi once told me the most important treatments were the ones that provided no data. Most of us spend most of our time trying to produce those special cause events (we just hope we created them through sound experimentation). Shewhart invented charts to help identify those given a rational series. Mahalanobis thought about this in the multivariate case. But the advice above is Fantastique. Merci beaucoup Veronique.

Re: Detection des points atypiques/ Outlier detection

VéroniqueB — Tue, 02 Feb 2021 14:50:32 GMT

Thank you for your reply. In fact, we relied on the tukey box plot but also on other compliant tests in connection with our outlier to confirm that this is indeed an atypical point due to the measurement issue.

Re: Detection des points atypiques/ Outlier detection

P_Bartell — Tue, 02 Feb 2021 21:54:03 GMT

@VéroniqueB In addition to the wise counsel of @Kevin_Anderson and @statman I suggest taking a look at the JMP Blog series authored by my former colleague (I'm a retired JMP Senior Systems Engineer) @JerryFish has and is composing on outliers. Here's a link to the third installment: Detecting Outliers Part 3 There are links to the first and second installment in the first paragraph of Part 3.