Banca de DEFESA: DOMINIQUE LOPES RAMOS

APPLICATION OF MACHINE LEARNING IN THE SEARCH FOR DARK MATTER AT LHC

Dark Matter, Machine Learning

Although the Standard Model (SM) is able to explain many phenomena in the world of particles, there are many problems that have not yet been clarified by this model. The nature of dark matter (DM), which makes up about 27% of the entire Universe, is an enigma both from the point of view of Particle Physics and Cosmology. DM does not interact with ordinary matter as we know it, so it cannot be detected in current experimental apparatus, being the perception of its existence only due to its gravitational interaction with ordinary matter. In the Large Hadron Collider (LHC), the search for DM is associated with its signature characterized by missing transverse energy ($\cancel{E}_T$) in experiments. Several theories beyond the SM consider a new DM candidate particle that can be a scalar, vectorial, a fermion of Dirac or Majorana particle. In this work, we consider a scalar DM candidate $\phi$ (with spin 0) predicted by ZP-TP-DM model that come from the decay of a new vector boson $Z'$ (spin 1) into a pair $T' \overline{T'}$ (called top fermionic partners with spin 1/2), giving rise to the final state $t\bar{t}\phi \phi$. Using Machine Learning (AM) techniques, we separate the background events ($pp \longrightarrow t\bar{t} Z $, with $ Z $ decaying in neutrinos and their respective antineutrinos) from the signal events ($ pp \longrightarrow T'\overline{T'} \longrightarrow t\bar{t} \phi\phi $) generated by CalcHEP. We built a deep neural network (DNN) to separate background and signal events and obtained values close to 1 for the Area Under the Receiver Operating Characteristic curve (AUC), indicating that the created classifier efficiently separated the events. However, the results obtained for statistical significance represent an ideal situation, since we did not include in the analysis the decay of the pair $t\bar{t}$.