Improvement of the Higgs Mass Resolution in the Decay Channel H -> bb in Associated Production with a W Boson at the ATLAS Detector Using Multivariate Regression

In the context of Higgs searches in the H ! bb¯ decay channel at ATLAS a good invariant di-jet mass resolution is important for the improvement of the signal to background ratio. This thesis demonstrates a method to improve the invariant mass resolution by using boosted decision trees. These decision trees were trained to give a better estimate of the true b-jet energy based on several input variables describing the jet. The influence of the input variables was investigated and several test were performed in order to optimize the setup that was used in this multivariate regression technique. As a result the invariant mass resolution of two b-jets was improved by 30% to a value of 10.9%. The multivariate regression technique was also successfully applied to other physics processes containing b-jets such as top-quark pair production. In addition, a comparison of the multivariate regression to other jet energy corrections shows that its performance is similar or even better.