Multivariate analysis methods for the optimisation of searches for new physics with leptons at Belle II

Abstract

This thesis will cover the use of multiple multivariate analysis (MVA) methods approaches for the identification of Leptons in the Belle II experiment. First, I will compare two machine learning techniques in terms of classification and performance, then I will focus on Lepton Flavour universality (LFU) violation in tau decays using multivariate analysis methods.LFU requires that the three charged lepton couple to the W bosons in the same way. This means that any ratio between the couplings ge, gμ and gτ , should equal the unity. By measuring those ratios, one can possibly reject the standard model of particle physics. By performing high precision measurements, which where done by the the Belle II detector, located at the SuperKEKB accelerator in Tsukuba, Japan, one can determine the couplings. Multivariate analysis methods help in the Signal/Background classification which leads to a bigger phase space and ultimately to a more precise determination of the coupling ratios.The investigation of LFU violation is done in τ → lν ̄νdecays using 3x1 prong τ -pair events at Belle II. The algorithm is trained on MC14 data and the purity and efficiency metric will then be compared to a similar analysis which uses a cut based approach.