Use of Machine Learning for gamma/hadron separation with HAWC

Abstract

Background showers triggered by hadrons represent over 99.9% of all particles arriving at groundbased gamma-ray observatories. An important stage in the data analysis of these observatories, therefore, is the removal of hadron-triggered showers. Currently, the High-Altitude Water Cherenkov (HAWC) gamma-ray observatory employs an algorithm based on a single cut in two variables, unlike other ground-based gamma-ray observatories (e.g. H.E.S.S., VERITAS), which employ a large number of variables to separate the primary particles. In this work, we explore machine learning techniques (Boosted Decision Trees and Neural Networks) to identify the primary particles detected by HAWC. Our new gamma/hadron separation techniques were tested on data from the Crab nebula, the standard reference in Very High Energy astronomy, showing an improvement compared to the standard HAWC background rejection method.

Gamma/Hadron Separation with the HAWC Observatory

“…The MLT models were trained on MC signal and MC background events. The MC simulation agrees with real data (both signal and background) for all the discrimination variables [32]. We chose to train with MC background because we obtained slightly worse MC testing results when training with real data 12 .…”

TeV Instrumentation: Current and Future

“…Deep learning methods are also starting to be exploited for WCD and SA instruments. Although machine learning methods have been shown to improve the performance of gamma/hadron separation, they have been fed so far only with a small selection of event parameters, rather than the full event information [58]. The treating of HAWC events as images in CNN has been tried as well, and has shown a gamma/hadron separation capability [94]; however, it is not clear yet if such a method improves the performance.…”

“…Machine learning methods using tree classifiers (such as random forest or boosted decision trees) are commonly used both in IACT [55][56][57] and WCD [58] data analysis to aggregate the information from different shower/image parameters into a single gamma/hadron separation parameter.…”

TeV Instrumentation: current and future

“…The deep learning methods are also starting to be exploited for WCD and SA instruments. While machine learning methods are shown to improve the performance of the gamma/hadron separation, they have been fed so far just with a small selection of event parameters, rather than the full event information [58]. Treating of HAWC events as images in CNN has been tried as well, and has shown a gamma/hadron separation capability [94], however it is not clear yet if such a method improves the performance.…”