Eliminating Cosmic Ray Backgrounds for SBND’s sub-GeV Dark Matter Search

In late Fall 2023 the SBND (Short-Baseline Near Detector) experiment on the Booster Neutrino Beam (BNB) at Fermilab will begin taking data. Fixed-target neutrino experiments can be used to probe the sub-GeV Dark Matter (DM) spectrum.1 Due to SBND’s proximity to the beam target, it will have access to very high statistics, including DM particles produced in the beam. DM scattering events are simulated using the BdNMC event generator.2 The DM events simulated in SBND are subsequently reconstructed using the Pandora pattern recognition suite of algorithms. In this work we focus on DM-e neutral current (NC) elastic scattering with no other hadronic behavior at the vertex. We next implement two models in order to differentiate between the cosmic and DM slices. We build both a traditional BDT, which was used in the original CRUMBS analysis and a more sophisticated nonlinear Neural Network (NN). Both were trained separately on distinctly optimized parameters to perform binary classification between the two types of models. We train both models on 20 epochs and use a 0.35 learning rate for the BDT, compared to a learning rate of 0.001 for the NN.