New gamma/hadron separation parameters for a neural network for HAWC

TL;DR

This paper introduces new gamma/hadron separation parameters for a neural network to enhance background suppression in the HAWC observatory, improving high-energy source detection but with some trade-offs at lower energies.

Contribution

The study identifies and integrates new separation parameters into a neural network, improving gamma-ray background suppression in HAWC data analysis.

Findings

Improved significance of high-energy gamma-ray sources.

Enhanced background suppression efficiency at TeV energies.

Slight performance decrease at GeV energies.

Abstract

The High-Altitude Water Cherenkov experiment (HAWC) observatory is located 4100 meters above sea level. HAWC is able to detect secondary particles from extensive air showers (EAS) initiated in the interaction of a primary particle (either a gamma or a charged cosmic ray) with the upper atmosphere. Because an overwhelming majority of EAS events are triggered by cosmic rays, background noise suppression plays an important role in the data analysis process of the HAWC observatory. Currently, HAWC uses cuts on two parameters (whose values depend on the spatial distribution and luminosity of an event) to separate gamma-ray events from background hadronic showers. In this work, a search for additional gamma-hadron separation parameters was conducted to improve the efficiency of the HAWC background suppression technique. The best-performing parameters were integrated to a feed-foward Multilayer Perceptron Neural Network (MLP-NN), along with the traditional parameters. Various iterations of MLP-NN's were trained on Monte Carlo data, and tested on Crab data. Preliminary results show that the addition of new parameters can improve the significance of the point source at high-energies (~ TeV), at the expense of slightly worse performance in conventional low-energy bins (~ GeV). Further work is underway to improve the efficiency of the neural network at low energies.