Electron Identification using Machine Learning in the MPD Experiment at NICA

TL;DR

This paper explores machine learning techniques, specifically MLP and BDT classifiers, to enhance electron identification in the MPD experiment at NICA, aiming to improve efficiency and purity over traditional methods.

Contribution

It introduces machine learning classifiers for electron identification in the MPD experiment, comparing their performance to traditional cut-based approaches.

Findings

MLP and BDT classifiers outperform traditional methods in efficiency.

Machine learning approaches maintain high purity in electron identification.

Momentum-dependent strategies improve classifier performance.

Abstract

We present studies of electron identification (eID) in the MPD experiment at NICA using machine learning techniques. The goal is to improve electron identification efficiency while preserving high purity, which is crucial for dielectron analyses. We compare electron identification performance between traditional cut-based approach and Machine learning. For machine learning based approach different classifiers, namely, Multi-Layer Perceptron (MLP) and Boosted Decision Tree (BDT) were trained with momentum-integrated and momentum-differential strategies using the \texttt{CERN ROOT TMVA} package.