Low-Energy Event Selection in IceCube using Machine Learning

Estimates of star formation rates in the Milky Way predict about three supernovae per century in the galaxy, of which two would be core-collapse supernovae (CCSNs). However, we have not observed a supernova in our galaxy in the last 400 years - a potential discrepancy that calls for novel methods of supernova observation. Neutrino signals from CCSNs could not only provide up to 24 hours advance warning of the explosion, but also help us study characteristics of neutrinos. The IceCube Neutrino Observatory, located at the South Pole, is currently the world's largest neutrino detector and is a part of the Supernova Early Warning System (SNEWS). In the event of a supernova, data from IceCube would be instrumental in alerting the world of an imminent supernova in our galaxy and the Magellanic Clouds. However, IceCube is largely sensitive to very high energy neutrinos (100 GeV-1 PeV), while supernova neutrinos usually lie in the 10 MeV range. This calls for the need to separate the desired signal from a variety of background events. Using existing simulations of neutrino events in the IceCube detector, including atmospheric neutrinos and background noise, machine learning algorithms can be developed and utilized to identify low-energy events using a variety of classification criteria.