A machine learning algorithm for energy reconstruction and binary classification of elastic and inelastic neutrino scattering events at the SND@LHC

This Bachelor Research Thesis (BTR) aims to improve the accuracy of energy reconstruction for particle showers within an energy range of 200-400 GeV passing through the Scintillating Fibre (SciFi) planes of the prospective Scattering and Neutrino Detector at the Large Hadron Collider (SND@LHC). To this end, the effects of adjusting several parameters of an algorithm using a Convolutional Neural Net (CNN) are investigated. The implementation of different loss functions significantly improved the accuracy of the algorithm and greatly reduced the overestimation of the predicted shower energy. This further prepares the algorithm for realistic implementation at the SND@LHC. Additionally, the BTR aims to create a classification algorithm for the identification of elastic and inelastic electron neutrino scattering events. The energy range of interest is 100 - 2000 GeV, which corresponds to the expected neutrino energy at the SND@LHC. An accuracy of 94.36% was achieved for the most realistic scenario. Moreover, the overall results provide a useful first insight into the (dis)advantages of using a CNN for particle shower classification. Further research is recommended to improve the algorithm’s ability identify particle showers that leave only a few hits in the detector.