Application of Machine Learning Algorithms for Searching BSM Higgs Bosons Decaying to a Pair of Bottom Quarks

Abstract

Although the discovery of the 125 GeV Higgs boson confirms the Higgs mechanism of the Standard Model (SM), many theories beyond the SM have been introduced to address several phenomena yet to be explained by the SM. For instance, the 2-Higgs Doublet Models is the simplest extension of the SM Higgs sector and predicting the existence of additional Higgs bosons at different states. The aim of this study is to search for machine learning (ML) algorithms which have been widely used in High Energy Physics. This will improve the sensitivity of the search for BSM Higgs bosons produced in association with a bottom quark () that then decays into a pair of bottom quarks (); the predominant decay channel of the Higgs boson, though, buried by a large multi-jet background process. In this study, we train 2 different ML algorithms: Tree-based models and Neural Networks, to classify signal and background events collected by the Compact Muon Solenoid detector from proton-proton collisions at 13 TeV. The evaluation metrics are calculated to provide classification efficiencies from different models. The results show that the classification of signal and background processes can be improved using ML techniques. Neural Networks reported the highest AUC score of 0.951 which is comparable with Adaptive Boosting model, while Decision Trees (DTs) and Random Forest models slightly underperformed by 2 - 3 %. We therefore can make use of the trained models as signal vs background classifiers to perform further statistical analysis searches for BSM Higgs bosons. GRAPHICAL ABSTRACT