Interpretable Machine Learning Model Development for Background Rejection in LEGEND

Neutrinoless Double Beta Decay (0νββ) is the hypothetical process which, if discovered, would shed light on persistent puzzles in the Standard Model. Detecting 0νββ is a major research interest in nuclear physics and is the primary task of the Large Enriched Germanium Experiment for Neutrinoless ββ Decay (LEGEND). LEGEND aims to utilize an array of High-Purity Germanium (HPGe) detectors with a novel inverted-coaxial point-contact (ICPC) design to detect 0νββ. Due to the requirement for unambiguous discovery of 0νββ, background rejection methods play a critical role in LEGEND. We have developed a Machine Learning algorithm to reject some of the most prominent backgrounds. Using an interpretable Boosted Decision Tree model, multiple pulse shape parameters can be analyzed simultaneously to improve rejection performance. Additionally, the interpretability of the model makes it a useful tool for discovering new correlations between parameters. By learning from the machine, we can gain a better understanding of detector microphysics.