Enhancing Electron Identification Near Jets Using Machine Learning in High-Energy Physics

In high-energy physics, precise electron identification is critical for accurate collision event reconstruction. This task becomes particularly challenging in busy environments where electrons are closely surrounded by complex structures and objects such as jets. Traditional methods, which employ a Likelihood-based approach using various observables and features, are generally designed to handle a wide range of scenarios. However, they are not specifically tailored for the intricate conditions present in jet-rich environments – characterized by dense overlapping signals and significant background noise– where their performance can be markedly suboptimal. To address this challenge, the project explores the application of machine learning (ML) techniques to improve electron identification near jets. By leveraging simulated data from high-energy collisions, we extract a variety of electron and jet observables, and employ supervised learning algorithms to build a classification model. The model aims to enhance the accuracy of distinguishing true electrons from wrongly reconstructed ones. Initial results indicate that the ML approach significantly improves performance compared to traditional methods in high-density jet environments. TThese results pave the way for more precise electron identification in the ATLAS experiment and other high-energy physics experiments. We achieved a substantial gain of 15% in identification efficiency compared to the ATLAS standard method, for electrons near jets. This improvement demonstrates that physics analyses involving electrons in dense environ- ments can benefit significantly, particularly from the increased statistics available when applying the ML techniques developed in this project.