Top squark signal significance enhancement by different machine learning algorithms

Abstract

In this paper, a study of four different machine learning (ML) algorithms is performed to determine the most suitable ML technique to disentangle a hypothetical supersymmetry (SUSY) signal from its corresponding Standard Model (SM) backgrounds and to establish their impact on signal significance. The study focuses on the production of SUSY top squark pairs (stops), in the mass range of [Formula: see text][Formula: see text]GeV, from proton–proton collisions with a center of mass energy of 13[Formula: see text]TeV and an integrated luminosity of [Formula: see text], emulating the data-taking conditions of the run II Large Hadron Collider (LHC) accelerator. In particular, the semileptonic channel is analyzed, corresponding to final states with a single isolated lepton (electron or muon), missing transverse energy, and four jets, with at least one tagged as [Formula: see text]-jet. The challenging compressed spectra region is targeted, where the stop decays mainly into a [Formula: see text] boson, a [Formula: see text]-jet, and a neutralino ([Formula: see text]), with a mass gap between the stop and the neutralino of about 150[Formula: see text]GeV. The ML algorithms are chosen to cover different mathematical implementations and features in ML. We compare the performance of a logistic regression (LR), a Random Forest (RF), an eXtreme Gradient Boosting, XGboost (XG) and a Neural Network (NN) algorithm. Our results indicate that XG and NN classifiers provide the highest improvements (over 17%) in signal significance, when compared to a standard analysis method based on sequential requirements of different kinematic variables. The improvement in signal significance provided by the NN increases up to 31% for the highest stop mass considered in this study (800[Formula: see text]GeV). The RF algorithm presents a smaller improvement that decreases with stop mass. On the other hand, the LR algorithm shows the worst performance in signal significance which even does not compete with the results obtained by an optimized cut and count method.