Development of a Machine Learning Based Analysis Chain for the Measurement of Atmospheric Muon Spectra with IceCube

Abstract

High-energy muons from air shower events detected in IceCube are selected using state of the art machine learning algorithms. Attributes to distinguish a HE-muon event from the background of low-energy muon bundles are selected using the mRMR algorithm and the events are classified by a random forest model. In a subsequent analysis step the obtained sample is used to reconstruct the atmospheric muon energy spectrum, using the unfolding software TRUEE. The reconstructed spectrum covers an energy range from $10^4\,$GeV to $10^6\,$GeV. The general analysis scheme is presented, including results using the first year of data taken with IceCube in its complete configuration with $86$ instrumented strings.