Evaluation of heavy flavour background for low mass dielectrons using machine learning techniques

Abstract

In Quantum Chromodynamics (QCD), chiral symmetry restoration is expected to occur at high temperatures and densities, such as those created in relativistic heavy-ion collisions. Low mass dielectrons (electron-positron pairs) serve as penetrating probes of the hot and dense medium created in such collisions, providing multiple insights of the phenomenon. This thesis focuses on the evaluation and reduction of the heavy-flavour background from proton-proton collisions at the ALICE experiment, using Machine Learning techniques and enabling a clearer view of the dielectron invariant mass spectrum. Our primary goal is to train a Machine Learning model, specifically using the XGBoost algorithm, with Monte Carlo simulated data. This model aims to effectively distinguish dielectrons originating from heavy-flavour hadron decays from those produced by other sources, while maintaining high accuracy and efficiency. We will also utilise previous works on proper dielectron track selection using Domain Adversarial Neural Networks (DANN). Our results indicate significant reduction at the low-mass dielectron yield, a result that still needs further validation with more Run 3 data and higher statistics in the Monte Carlo simulations. Still, we observe promising indications that our model is effectively working, as small peaks appear in the resulting invariant mass ranges for the ρ, ω, ϕ and J/ψ mesons.