Enhancing detectablility of tau-sneutrino signatures using machine learning

Abstract

In this thesis, the collider signatures of the scenario with a tau-sneutrino next-to-lightest supersymmetric particle (NLSP) at LHC are studied using machine learning. The parameter region of the non-universal Higgs masses model, where the tau-sneutrino is the NLSP, is studied to find a parameter point which satisfies constraints from recent experimental results. We look at the tri-lepton signature from two same sign hadronic taus and a muon. This signature have its main contribution from the slepton and sneutrino pair production channel. The aim is to enhance detectability of this signature by using a deep neural network trained on monte carlo simulated collider events. The best performing deep neural network is a multi class classifier, which is compared to other neural network architectures and a boosted decision tree. The required integrated luminosity for a 5σ significance discovery using √s=13 TeV is found to be L(5σ)= (3.4 ±0.7)⨉10³ 1/fb. We find that the multi class deep neural network performs better by a factor of 2.0 than the traditional optimized cuts.