Search for tbH$^+(\tau\tau)$ with Performance Optimisation for Signal and Background Separation Using Machine Learning with ATLAS Data

The search for charged Higgs bosons, predicted by the Two Doublet Higgs Model and the Minimal Supersymmetric extension of the Standard Model, is challenging because of a large number of background processes and the unknown mass of the charged Higgs bosons. This thesis proposes to use machine learning to separate signal tbH$^+ \rightarrow$ tbW$\tau\tau$ from tth, ttW , ttZ , tt, VV and other background processes. A multi-model approach is proposed, where each model is sensitive in a certain mass range to achieve large significance in its dedicated mass section. Four different model types are optimized and the best model is selected for each mass of the charged Higgs boson analysis. Permutation feature ranking is used for each best model to determine the most important features. Based on the highest-ranking features, feature reduction is demonstrated to reduce the sensitivity only slightly. Results are expressed as expected 95% CL limits.