I have mentioned several times that the Hubble classification is a traditional approach of classification that is not adapted to modern data. I have also noticed that in the astrophysical literature modern statistical tools of clustering are often used in the sole purpose of retrieving the Hubble sequence without using the eyes. Or course, this rarely works, and I have found a very nice illustration of this pitfall.
Neichel et al (Astronomy and Astrophysics 2008, 484, 159) introduced a decision tree to classify galaxies from various observables, most of them quantitative and objective :
Delgado-Serrano in his thesis in 2010 presents a more complete version of this decision tree :
Humm, do you see what I see? From Wikipedia, I learn that “a decision tree has only burst nodes (splitting paths) but no sink nodes (converging paths)”.
I can’t help but see in these decisions trees a strong will to enforce the Hubble classification by putting together objects that should not be gathered together. It seems to me that the above tree must be drawn in the following way:
Yes, many more species appear. It is up to you to force for instance Pec/Irr 2 to be grouped with Pec/Irr 8, but by doing so you may gather oranges and bananas in the same basket. Ok if you want to eat fruits, but for a cake or a cocktail the difference matters…
So my question is how long will astronomers stick to an obsolete classification and continue spending their time to squander very informative data and sophisticated analyses by forcefully fitting their results into a taxonomy based on one unique and subjective descriptor?
Astrocladistics 