Use of Machine Learning Methodologies in the PROSPECT Experiment

The Precision Reactor Oscillation and Spectrum Experiment (PROSPECT) measured electron antineutrinos (𝜈̄ₑ) emitted by the High Flux Isotope Reactor in 2018 to search for possible oscillation signatures from sterile neutrinos. These 𝜈̄ₑ were detected via the inverse beta decay (IBD) interaction, identified by a distinctive double-coincidence signature that discriminates signal from background. Despite challenges posed by photomultiplier tube (PMT) base failures in some detector segments, innovative machine learning techniques significantly enhanced position and energy reconstruction, as well as particle classification—particularly in single-PMT segments where traditional reconstruction was limited. This work demonstrates the effectiveness of convolutional neural networks and graph convolutional networks in improving data analysis, resulting in a 3.3% increase in effective statistics compared to conventional event selection methods. These results highlight the potential of machine learning to boost performance in PROSPECT and other segmented particle detectors, underscoring its broad applicability in neutrino physics and beyond.