Comparative pulse shape discrimination study for Ca(Br, I)$_2$ scintillators using machine learning and conventional methods

TL;DR

This study compares machine learning and traditional methods for pulse shape discrimination in calcium halide scintillators, demonstrating that machine learning improves background rejection, especially at low energies.

Contribution

It introduces the application of ML techniques to PSD in calcium halide scintillators and shows their superior performance over conventional methods.

Findings

ML methods outperform conventional PSD methods across all scintillators.

Machine learning is more effective at low energies for scintillators with low light output.

ML techniques enhance background rejection in particle physics experiments.

Abstract

In particle physics experiments, pulse shape discrimination (PSD) is a powerful tool for eliminating the major background from signals. However, the analysis methods have been a bottleneck to improving PSD performance. In this study, two machine learning methods -- multilayer perceptron and convolutional neural network -- were applied to PSD, and their PSD performance was compared with that of conventional analysis methods. Three calcium-based halide scintillators were grown using the vertical Bridgman--Stockbarger method and used for the evaluation of PSD. Compared with conventional analysis methods, the machine learning methods achieved better PSD performance for all the scintillators. For scintillators with low light output, the machine learning methods were more effective for PSD accuracy than the conventional methods in the low-energy region.