| dc.description.abstract | The use of machine-learning techniques in particle physics is relatively new and unexplored when compared to more conventional methods that use rectangular cuts. This thesis has its focus on the reconstruction of D± mesons in transverse momentum range 2 ≤ pT ≤ 24 GeV/c, that are produced in proton-proton collisions at centre-of-mass energy √s = 5.02 TeV with ALICE. The goal of this thesis is to try to reduce the current statistical uncertainties on the extraction of the D± raw yield by leveraging boosted decision trees for signal selection. This is especially important because, besides being an excellent test for (perturbative) quantum chromodynamics, this type of measurement constitutes the baseline for the corresponding measurement in heavy-ion collisions. Currently the uncertainties on the baseline measurement are a limiting factor when comparing to models, to shed light on the in-medium energy loss of charm quarks in the quark-gluon plasma. A precise estimation of the baseline helps with extracting key plasma properties and therefore it contributes to paint a picture of the early Universe, only a few microseconds after the Big Bang, since it is believed that at that point all the matter in the Universe was in the quark-gluon plasma phase. The results presented in this thesis show a promising increase in significances, which directly correlates to an improved statistical uncertainty, compared to those obtained from the topological selection cuts used for the standard ALICE analysis, along with some interesting deviations in efficiency-corrected yields. The stability of the results obtained from the use of boosted decision trees is shown to be high, especially at mid-to-high pT. | |
