Speaker
Anton Lukyanov
(PhD student)
Description
One of main tasks in PAMELA experiment is identification of cosmic ray (CR) antiprotons on different
kinds of background. One example of such background at low energies are pi-meson particles which are
generated in elements of spectrometer construction by high-energy CR protons. We propose an approach
based on machine learning methods, in particular, Support Vector Machines (SVM). We use two
different sets of features for classification: track system features (12 measurements of ionization
energy losses along particle track) and calorimeter features (different combinations of energy
release inside calorimeter along reconstructed particle trajectory).
Constructed classifier showed classification accuracy of 96% for antiprotons and 89% for pi-mesons
when rigidity $R$ is up to 2 GeV. When $R \in (2, 5]$ classification accuracy for pi-mesons is 15%.
For evaluation of classification accuracy we used k-fold cross-validation method with $k=5$.
Primary author
Anton Lukyanov
(PhD student)
Co-authors
Andrey Mayorov
(National Research Nuclear University MEPhI)
