An Efficient Boosted Decision Tree for Background Rejection in the Light Dark Matter Experiment (LDMX)

The Light Dark Matter eXperiment (LDMX) is proposed to use the SLAC LCLS-II accelerator for 8 GeV electrons on target to search for evidence of dark matter production in the sub-GeV mass range. From the experiment’s high granularity Electromagnetic Calorimeter (ECal) information, discriminating variables are calculated and fed to a Boosted Decision Tree (BDT) trained to distinguish between signal and background events. The ECal plays a critical role in both the trigger and offline photon veto, reducing difficult photonuclear and other backgrounds by many orders of magnitude while retaining high efficiency for signal. Previously, constructing tracks from Minimum Ionizing Particles (MIPs) for every event added significant computational cost. A new BDT approach eliminates the need for full MIP tracking, applying it only to the small fraction of events that pass earlier stages, resulting in a substantial reduction in processing time. The study presented in this talk uses a detailed material description of the detector to produce fully simulated events with realistic tracking of the electron in the Recoil Tracker. We demonstrate the new cutflow’s effectiveness in rejecting difficult photonuclear and electronuclear backgrounds.