Measurement of the B(W->cq)/B(W->qq') decay branching fraction ratio with the CMS detector at the LHC and Uncertainty estimation of machine learning models in particle physics

Particle physics aims to understand the fundamental structure of matter and its interactions, relying on precise experimental measurements for its progress. The Standard Model (SM) provides the theoretical foundation for describing elementary particles and their interactions. The Large Hadron Collider and its experiments, particularly CMS, enable precise tests of the SM. This thesis focuses on testing weak universality within the SM, proposed for quarks by Cabibbo and extended by Kobayashi and Maskawa, to explain transitions between up-type and down-type quarks via weak interactions. The probability of these transitions, encoded in the CKM matrix, is determined experimentally. This thesis presents a measurement of the charm quark production rate in W decays relative to other quark flavors, quantified by the branching fraction ratio \Rcw = \BRWtoC. According to SM, \Rcw is expected to be 1/2, making its measurement a direct test of CKM unitarity and weak universality. The analysis utilizes a large sample of W bosons produced in \ttbar events, where one of the W bosons decays leptonically and the other hadronically. Charm quark identification, or "charm tagging," is crucial for this measurement and involves identifying a muon within jets originating from charm hadron decays. This technique provides a clean sample with well-controlled systematics, enabling a more precise determination of \Rcw. The measurement presented reduces the uncertainty of the current world average by approximately half, marking a significant advancement in precision tests of the SM. Furthermore, this thesis explores uncertainty estimation in particle physics using machine learning techniques. By applying Bayesian neural networks, probabilistic random forests, and local ensembles to CMS open data, this work underscores the importance of reliable uncertainty estimation in enhancing the robustness of analyses using modern computational tools.