Muon and Pion Identification at BESIII Based on Machine Learning Algorithm

Abstract

BESIII is designed to study physics in the τ-charm energy region utilizing the high luminosity BEPCII. For collision physics experiments like the BESIII experiment, particle identification (PID) is one of the most important and commonly used tools for physics analysis. The effective µ/π identification performance is of great significance for most of BESIII physics analysis. However, due to the close masses of these two particles, as well as the intrinsic correlation between multiple detector information, traditional methods at BESIII is facing challenges in µ/π identification. In recent decades, machine learning (ML) techniques have been rapidly developed and have shown successful applications in HEP experiments. The PID based on ML provides powerful capability of combining more detection information from all sub-detectors with the data-driven approach. In this article, targeting at the µ/π identification problem at the BESIII experiment, we have developed a new PID algorithm based on the gradient boosted decision tree (BDT) model. Preliminary results show that the XGBoost classifier provides obviously higher discrimination power than traditional methods. In addition, based on the substantial amount of high-quality data taken by the BESIII detector, a method of evaluating and suppressing the systematical error of the ML model is also introduced, which is critical for applying the model to physics studies.

Keywords

• particle identification
• machine learning
• boosted decision tree
• besiii