Higgs production channel classification in searches for long-lived dark photon jet decays with the ATLAS detector using machine learning

Abstract

In studies looking for long-lived dark photons that produce displaced jet signatures in the ATLAS detector, analyses depend on cuts to ensure orthogonality between production modes of the Higgs boson. This thesis proposes the use of a trained machine learning model to replace these restrictive cuts. Three models are trained to classify events coming from ggF and VBF production modes and one is selected to replace currently used cuts. The selected model reaches a 92.2% accuracy on the VBF samples, an increase of 12 times the accuracy obtained from cuts; and a 97.8% accuracy for the ggF samples, statistically the same accuracy obtained from cuts. Using this model, orthogonality between VBF and ggF studies is assured while maintaining a large portion of the signal data available for analysis.