Identification of muonic decays of tau pairs at the Belle II experiment through the implementation of machine learning algorithms

Abstract

The Belle II experiment, installed at the electron-positron collider SuperKEKB in Tsukuba, Japan, plans to collect around 50/ab of data over the course of its lifetime; around 50 times more than its predecessor Belle. This presents a unique opportunity to study heavy lepton decays with unmatched precision. The aim of this thesis is to implement and test machine learning algorithms for the identification of muonic decays of tau pairs: e+e− τ+τ− μ+(νμντ )μ−(νμντ ).Machine Learning algorithms, which are increasingly popular in many areas of scientific research, are well suited for the analysis of large data sets. A method is trained on a set of data from which it learns to extract important features. Applied on independent data, it then tries to distinguish the signal from the background. The hypothesis of this thesis is, that these algorithms outperform humans in doing so. Different algorithms were trained on monte-carlo simulated data and then compared to one another and to classical cut-based analysis. The best performing algorithm was then used to calculate the cross-section of the process e+e− τ+τ− on an independent, also simulated data set.The results showed a superior performance of the Machine Learning models over cut- based analysis and a more accurate calculated cross-section. This suggests that these algorithms are indeed better at separating signal from background events than humans, at least in the context of the decays investigated here.