Astrophysical or Terrestrial: Machine Learning Classification of Gravitational-wave Candidates Using Multiple-search Information

Abstract Low-latency gravitational-wave alerts provide the greater multi-messenger community with information about the candidate events detected by the International Gravitational-Wave Network. Prompt release of data products such as the sky localization, false alarm rate, and p astro values allow astronomers to make informed decisions on which candidate gravitational-wave events merit target of opportunity follow-up. However, false alarms where a gravitational-wave candidate, or trigger, is the result of terrestrial noise, are an inherent part of gravitational-wave searches. In addition, with the presence of multiple gravitational-wave searches, different searches may have varying assessments of the significance of a given trigger. As a complement to quantities such as p astro , we provide a Machine Learning (ML) based approach to determining whether candidate events are astrophysical or terrestrial in nature. This ML classifier utilizes information provided by multiple low-latency search pipelines and downstream data products in its feature space, and is trained on a realistic real-time simulation campaign called the Mock Data Challenge (MDC). The Random Forest implementation has an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.96 and accuracy of 0.90 on the MDC training set compared to p astro which had an AUC of 0.72 and an accuracy of 0.75. We tested on events from Advanced LIGO (aLIGO)’s, Advanced Virgo (AdVirgo)’s, and KAmioka GRAvitational-wave observatory (KAGRA)’s third observing run (O3) and got an AUC of 0.93 and accuracy of 0.86 compared to p astro which had an AUC of 0.82 and accuracy of 0.75.