Boosted decision tree-driven cut-based analysis for event recognition in Stealth SUSY decays for √s = 13 TeV
Boosted Decision Trees (BDT) algorithm using the XGBoost toolkit was applied to a dataset that contains a Stealth SUSY decay chain signal and QCD multijets background. Using the top ranking features according to BDT, the following cuts greatly reduced the amount of QCD multijets background in the dataset: the number of large-radius (R = 1.0) jets in an event must be greater than 3 and the angular separation distance between the leading and next-leading photon (ΔRγ1,γ2) must be greater than 0.10. The calculated signal significance Z = s/√(s + b) for this search at benchmark point (m~q = 1650 GeV/m~χ01 = 250 GeV) scaled at 36 fb-1 integrated luminosity is 4.76; higher than the 1.48 significance of the latest ATLAS search that probes SUSY processes for the same benchmark point and integrated luminosity [Phys. Rev. D 97, 092006 (2018)].