Speakers
Description
Particle-identification (PID) is a crucial part for all analyses performed with the KM3NeT neutrino telescopes. It is used to separate atmospheric neutrino from background events (atmospheric muons, $ ^{40}\mathrm{K}$ decay) and further, to distinguish track- and shower-like events, allowing the identification of the neutrino flavour.
One of the main algorithms used for event classification in KM3NeT makes use of Boosted Decision Trees (BDTs), that are trained on a wide set of parameters. In this contribution, the used training features are presented, such as variables from the event reconstruction and likelihood distributions of expected signals in the detector, as well as the process of extending the list by the evaluation of the benefit of new features towards the PID efficiency. The procedure by which the classifiers are trained and included into the data analysis will be also discussed.
With only 6 detection units of the ORCA detector, it will be shown that a separation between track- and shower-events can be achieved with high accuracy at energies above $10 \ \mathrm{GeV}$. Future improvements of the track and shower reconstruction algorithms, together with the quickly growing instrumented volume of ORCA, are expected to bring a higher track-shower separation power at energies lower than that. Moreover, the current muon and neutrino event classification will potentially allow to reduce the atmospheric muon contamination in the event samples, down to the few-percent level. Additionally, our study promises a high rejection efficiency of background events induced by $ ^{40}\mathrm{K}$ decay.
| Collaboration | KM3NeT |
|---|
