Machine Learning-Based Tagging of SiPM Crosstalk in LEGEND-200.

LEGEND is a Ge-76 based search aiming to discover whether neutrinos are Majorana in nature. This would provide a potential explanation for the observed matter-antimatter asymmetry. To reject backgrounds caused by external gamma sources, LEGEND employs a liquid argon light readout system instrumented with silicon photomultipliers (SiPMs). Currently, certain forms of crosstalk between the SiPMs and the germanium detectors are being missed by traditional data tagging methods, leading to low 0vββ signal efficiency if liquid argon trigger thresholds are lowered. To address this, a machine learning based tag was developed through the use of unsupervised affinity propagation (AP) and a supervised support vector machine (SVM). The resulting model has demonstrated the ability to tag crosstalk not previously detected by traditional methods.