Improved $\gamma$/hadron separation for the detection of faint gamma-ray sources using boosted decision trees

Imaging atmospheric Cherenkov telescopes record an enormous number of cosmic-ray background events. Suppressing these background events while retaining $\gamma$-rays is key to achieving good sensitivity to faint $\gamma$-ray sources. The differentiation between signal and background events can be accomplished using machine learning algorithms, which are already used in various fields of physics. Multivariate analyses combine several variables into a single variable that indicates the degree to which an event is $\gamma$-ray-like or cosmic-ray-like. In this paper we will focus on the use of boosted decision trees for $\gamma$/hadron separation. We apply the method to data from the Very Energetic Radiation Imaging Telescope Array System (VERITAS), and demonstrate an improved sensitivity compared to the VERITAS standard analysis.