How to use Machine Learning to improve the discrimination between signal and background at particle colliders

TL;DR

This paper evaluates various machine learning algorithms and libraries for improving signal-background discrimination in particle collider experiments, providing practical guidelines for analysts to optimize their use cases.

Contribution

It compares the performance of multiple ML algorithms and libraries, including BDTs and Neural Networks, in simulated LHC data, offering new insights and recommendations.

Findings

Neural Networks outperform Boosted Decision Trees in certain scenarios.

Toolkit for Multivariate Data Analysis with ROOT is effective for quick assessments.

Guidelines are provided for choosing algorithms based on data features and size.

Abstract

The popularity of Machine Learning (ML) has been increasing in the last decades in almost every area, being the commercial and scientific fields the most notorious ones. Concerning particle physics, ML has been proved as a useful resource to make the most of projects such as the Large Hadron Collider (LHC). The main advantage provided by ML is reducing the time and effort put into the measurements done by experiments, while improving the performance. With this work we aim to encourage scientists at particle colliders to use ML and to try the different alternatives we have available nowadays, focusing in the separation between signal and background. We assess some of the most used libraries in the field, like Toolkit for Multivariate Data Analysis with ROOT, and also newer and more sophisticated options like PyTorch and Keras. We also check how optimal are some of the most common algorithms for signal-background discrimination, such as Boosted Decision Trees, and propose the use of others, namely Neural Networks. We compare the overall performance of different algorithms and libraries in simulated LHC data and produce some guidelines to help analysts deal with different situations. Examples are the use of low or high-level features from particle detectors or the amount of statistics available for training the algorithms.