Gamma/Hadron Separation with the HAWC Observatory

TL;DR

This paper presents advanced machine learning techniques to improve gamma/hadron separation in the HAWC observatory, significantly enhancing the identification of gamma-ray events amid a large background of cosmic-ray hadrons.

Contribution

It introduces machine learning methods like boosted decision trees and neural networks for gamma/hadron separation, surpassing traditional cut-based techniques in HAWC data analysis.

Findings

Improved gamma/hadron separation efficiency.

Enhanced sensitivity to gamma-ray sources.

Reduction in background contamination.

Abstract

The High Altitude Water Cherenkov (HAWC) gamma-ray observatory observes atmospheric showers produced by incident gamma rays and cosmic rays with energy from 300 GeV to more than 100 TeV. A crucial phase in analyzing gamma-ray sources using ground-based gamma-ray detectors like HAWC is to identify the showers produced by gamma rays or hadrons. The HAWC observatory records roughly 25,000 events per second, with hadrons representing the vast majority ($>99.9\%$) of these events. The standard gamma/hadron separation technique in HAWC uses a simple rectangular cut involving only two parameters. This work describes the implementation of more sophisticated gamma/hadron separation techniques, via machine learning methods (boosted decision trees and neural networks), and summarizes the resulting improvements in gamma/hadron separation obtained in HAWC.