Machine Learning Application In Data Analysis In The Askaryan Radio Array Experiment

Askaryan Radio Array (ARA) is an ultra-high energy neutrino detection experiment that has been collecting data at the South Pole Station in Antarctica for a decade. To increase sensitivity to neutrino events, ARA is collecting high data volumes at low thresholds, and consequently its data set is dominated by background events such as thermal noise or noise from anthropogenic events. In this contribution, we build Machine Learning classification tools that are capable of discriminating background events from the expected neutrino candidates at the level of 1:10,000 or better. The methods employed to train event classifiers include Convolutional Neural Networks, Deep Neural Networks, and Boosted Decision Trees, with the inputs being images constructed from the event records. The classifiers are trained on background-dominated samples of real data and simulated neutrino candidates. The performance of these methods including the signal detection efficiencies, background rejections, and receiver operating characteristic curves is presented, with the comparison of the methods discussed.