Multiclass Classification Of Leptons In Proton-Proton Collisions At √s=13 TeV Using Machine Learning

Abstract

The Standard Model (SM) in particle physics does not explain everything we observe in the Universe. The non-zero mass of the neutrino is one such observation. The Inverse seesaw mechanism tries to explain the neutrino mass by with the existence of heavy neutrino masses and right-handed neutrinos. This leads to trilepton final states plus a neutrino from the decay of a W boson. We will use two types of neutrino signals with neutrino masses of 150 GeV and 450 GeV with data from proton-proton collisions collected by the ATLAS detector at sqrt{s}=13 TeV. We use machine learning (ML) techniques to train on these signals and use multiclass classification to classify lepton vertex permutations. We find the best performing multiclass classification model and use it to classify simulated backgrounds and signals. With the lepton vertices we study the charges and flavors of the leptons in the production and decay of the heavy neutrino. We find that the ML models predict many more events for SF than DF, giving different degrees of lepton flavor violation for the vertex permutations of the backgrounds. The significance of the 450 GeV signal was found to be higher compared to the 150 GeV signal, and higher for the invariant mass of the trilepton system compared to the missing transverse energy. The multiclass classification of the lepton vertices yield better performances than a simple benchmark analysis.