Likelihood and Deep Learning Analysis of the electron neutrino event sample at Intermediate Water Cherenkov Detector (IWCD) of the Hyper-Kamiokande experiment

TL;DR

This paper presents a machine learning framework to improve the identification of electron neutrino events in the Hyper-Kamiokande experiment's IWCD detector, aiming to reduce background contamination and enhance measurement precision.

Contribution

It introduces a novel machine learning approach for neutrino event classification, surpassing traditional likelihood-based methods in accuracy and efficiency.

Findings

ML framework improves event purity and efficiency

Enhanced background rejection in neutrino detection

Potential for more precise neutrino oscillation measurements

Abstract

Hyper-Kamiokande (Hyper-K) is a next-generation long baseline neutrino experiment. One of its primary physics goals is to measure neutrino oscillation parameters precisely, including the Dirac CP violating phase. As conventional $\nu_{\mu}$ beam generates from the J-PARC neutrino baseline contains only 1.5$\%$ of $\nu_{e}$ interaction of total, it is challenging to measure $\nu_{e}/\bar{\nu}_{e}$ scattering cross-section on nuclei. To reduce these systematic uncertainties, IWCD will be built to study neutrino interaction rates with higher precision. Simulated data comprise $\nu_{e}CC0\pi$ as the main signal with NC$\pi^{0}$ and $\nu_{\mu}CC$ are major background events. To reduce the backgrounds initially, a log-likelihood-based reconstruction algorithm to select candidate events was used. However, this method sometimes struggles to distinguish $\pi^{0}$ events properly from electron-like events. Thus, a Machine Learning-based framework has been developed and implemented to enhance the purity and efficiency of $\nu_{e}$ events.