Methods of machine learning for the analysis of cosmic rays mass composition with the KASCADE experiment data

We study the problem of reconstruction of high-energy cosmic rays mass composition from the experimental data of extensive air showers. We develop several machine learning methods for the reconstruction of energy spectra of separate primary nuclei at energies 1-100 PeV, using the public data and Monte-Carlo simulations of the KASCADE experiment from the KCDC platform. We estimate the uncertainties of our methods, including the unfolding procedure, and show that the overall accuracy exceeds that of the method used in the original studies of the KASCADE experiment.