Machine Learning on simulated forward calorimeter upgrade for JEF

Abstract

This thesis centers around using Machine Learning to distinguish true photons from hadronic split-offs detected by the JEF and GlueX experiments. The FCAL is an electromagnetic detector undergoing an upgrade to its inner region for improving neutral particle identification for the JEF project. Search for exotic η and η′ decays are the focal point JEF experiment as they test the limits of the current Standard Model and provide unique opportunities to search for Dark Matter candidates. The aim of this project is to develop of a shower classification scheme that can confidently separate between true photons and split-offs showers in the FCAL2 detector, as an extension of the study done on the original FCAL. Using simulated ω → π+π−π0 data, multiple Machine Learning models are trained and tested in their separation capabilities. The MLP and BDT models are found to provide robust classification. Their performance is tested in separate areas of the FCAL2, and validated on η → π+π−π0 to ensure application of these models to the JEF. They were both found to have consistently strong separation capabilities in all areas tested. This thesis outlines the motivation for JEF, details the ongoing detector upgrade, and explains the machine learning efforts on improve particle classification for the FCAL2.