What drives the choice of IQR vs MAD?

SamGG commented 5 months ago

Hi,

I don't catch the rational of IQR, especially because it is not adjusted when comparing to MAD. Could you explain me what IQR brings over MAD?

> x = rnorm(100000)
> mad(x)
[1] 1.003071
> IQR(x)
[1] 1.352785

giannimonaco commented 5 months ago

Hi, IQR is just less sensitive as it has a higher threshold when detecting the anomalies. It does not really bring anything else..

On Wed, Mar 27, 2024 at 5:59 PM Samuel Granjeaud @.***> wrote:

Hi,

I don't catch the rational of IQR, especially because it is not adjusted when comparing to MAD. Could you explain me what IQR brings over MAD?

x = rnorm(100000) mad(x) [1] 1.003071 IQR(x) [1] 1.352785

— Reply to this email directly, view it on GitHub https://github.com/giannimonaco/flowAI/issues/19, or unsubscribe https://github.com/notifications/unsubscribe-auth/AC2UTEDO7OCJVAYDF2ZWQ2DY2MXK5AVCNFSM6AAAAABFLWQBVOVHI2DSMVQWIX3LMV43ASLTON2WKOZSGIYTCOJXGU4TGMA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

SamGG commented 5 months ago

Thanks for your answer. So, IIUC, it is just about changing the level of significance. IMHO, alpha is a better choice for tuning. #20

giannimonaco / flowAI

What drives the choice of IQR vs MAD? #19