An Application of Machine Learning Techniques to Energy Reconstruction in IceTop Cosmic Ray Events

In order to understand the origins of cosmic rays and their interactions, determining their energies is essential. While the cosmic ray energy estimation with IceTop achieves adequate precision, its accuracy is affected by systematic uncertainties arising from hadronic-interaction models, different-mass primaries, and shower-to-shower fluctuations. In this contribution, we benchmark the performance of machine learning models for energy reconstruction using Monte-Carlo simulations, structured with features including arrival time of the signals, signal charges detected, and arrival direction of the air shower. We test four different element groups on these models: proton, helium, oxygen, and iron. The measured reconstruction accuracy is quantified by measuring the correlation coefficients and mean squared error of true and reconstructed energies. We will demonstrate how each element group performs individually and combined.