Improving electron/proton discrimination at high energies with CALET on the International Space Station

The CALorimetric Electron Telescope (CALET), operating aboard the International Space Station since October 2015, is an experiment dedicated to high-energy astroparticle physics. The primary scientific goal of the experiment is the measurement of the electron+positron flux up to the multi-TeV region. At such high energies, proton contamination - coupled with limited statistics - is the main challenge for this measurement and good electron/proton discrimination can be carried out by using machine learning techniques. So far, we have tested and used only algorithms implemented in the ROOT TMVA package: in particular, the Boosted Decision Tree (BDT) algorithm leads to proton contamination below 10% up to 7.5 TeV with an 80% electron efficiency. In principle, better performance can be achieved by using Python packages, which offer a larger variety of machine learning algorithms and tuning parameters compared to TMVA. In this work, we will present a comparison of the performance obtained with the BDT algorithm implemented in TMVA and Python (XGBoost), while alternative approaches based on neural networks (e.g., Keras) will be explored in future studies.