Study of a Multivariate Approach for the Background Rejection in the Scattering of Two Like-Charge $W^{\pm}$ Bosons with the ATLAS Detector at the LHC

This thesis presents the study of a multivariate approach for the background rejection in the scattering of two like-charge $W^{\pm}$ bosons with the ATLAS detector at the Large Hadron Collider. The scattering process can be accessed through the measurement of purely electroweak production of two like-charge $W^{\pm}$ bosons and two jets in the fully leptonic decay channel of the $W^{\pm}$ bosons. Although the characteristic signature of the final state of this production process already reduces most Standard Model backgrounds, other processes exist that leave the same experimental signature in the detector. QCD-initiated production of a $W^{\pm}$ boson and a $Z$ boson in association with two jets with leptonic decay of the $W^{\pm}$ and the $Z$ boson accounts for the largest background contribution. Thus, the focus of this thesis is set on the rejection of this background. As a very promising technique for this classification problem, boosted decision trees are studied in this thesis. The variable ranking of the NeuroBayes package is used to select an optimal set of input variables. Furthermore, the impact of various parameters of the boosted decision tree is investigated. Simulated data, mimicking the real data output of the ATLAS detector for a center-of-mass energy of 13 TeV, is used to train the boosted decision tree and to evaluate its performance. Based on these studies, a significance of 5.2 standard deviations is expected for 3.2 fb$^{-1}$ of proton--proton collision data for the discrimination between purely electroweak production of two like-charge $W^{\pm}$ bosons and two jets and QCD-initiated production of a $W^{\pm}$ boson and a $Z$ boson in association with two jets, neglecting systematic uncertainties. Compared to a cut-based benchmark analysis, this corresponds to a significance increase of about 47 %.