The problem of classifying elements of a data set as belonging to one class or another, depending on their characteristics, is a very, very well-studied one, and one which is particularly important in particle physics.

Imagine, for instance, that you collect events with four high-transverse-momentum leptons (electrons or muons) with the ATLAS or CMS detector, and you wish to sort out which of these fit better to the hypothesis of being originated by Higgs boson decay into two Z bosons (with each Z boson in turn producing a lepton pair) rather than to the alternative hypothesis of being due to the incoherent production of a pair of Z bosons -a process that has nothing to do with Higgs bosons. This means you need to classify the data events using their observed features.

Statistics data analysis is one of those things that experimental physicists learn along the way. It is not a topic usually included in the curriculum studiorum of physics students at Universities: only few basic ingredients are taught during laboratory courses, and not much is added to that during a typical Ph.D. program.

One usually learns the most common tools to fit histograms, combine measurements, estimate uncertainties on the field, as these things are always needed to produce publishable physics results. But several key statistical concepts often remain fuzzy and obscure in the mind of a large fraction of experimental physicists throughout their career. I know this because this happened to me, too - for quite a few years after my graduation.