Searches for the BSM scenarios at the LHC using decision tree based machine learning algorithms - A comparative study and review of Random Forest, AdaBoost, XGboost and LightGBM frameworks

Aug 5, 2026

Upload by Asher

Machine learning algorithms are used to separate rare signals from overwhelming backgrounds in high energy physics. This work surveys decision tree methods in Standard Model and Supersymmetry contexts, summarizing machine learning foundations and the operating principles of Random Forest, AdaBoost, XGBoost, and LightGBM. A case study on electroweakino production at the high-luminosity LHC shows improved search sensitivity over traditional cut-based approaches in compressed and non-compressed R-parity conserving SUSY scenarios, including hyperparameter optimization and SHapley-value feature-importance analysis.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages62

Reading Time2 h 36 min

AccessFree to download

Introduction

Basic concepts of machine learning

Loss Function

Overfitting and underfitting

Measures of classification performance

Machine learning in High Energy Physics

Signal and background events classification

Searches for RPC SUSY scenarios using BDT

Searches for RPV SUSY scenarios using BDT

Decision Tree algorithms

Random Forest

AdaBoost

XGBoost

LightGBM

Performance of different Decision Tree based algorithms-a RPC SUSY case study at the HL-LHC

Cut-and-count analysis

Machine Learning based analysis

Hyperparameter variation for different algorithms

Feature importance with SHapley

Comparison of results coming from different algorithms

Summary

arXiv :2405 .06040v1 [hep-ph] 9 May 2024 Searches for the BSM scenarios at the LHC using decision tree based machine learning algorithms: A comparative study and review of Random Forest, Adaboost, XGboost and LightGBM frameworks Arghya Choudhury, Arpita Mondal, and Subhadeep Sarkar Department of Physics, Indian Institute of Technology Patna, Bihar - 801106, India E-mail: [arghya@iitp. ac. in](arghya@iitp. ac. in), arpita [1921ph15@iitp. ac. in](1921ph15@iitp. ac. in), subhadeep [1921ph21@iitp. ac. in](1921ph21@iitp. ac. in) Abstract: Machine learning algorithms are now being extensively used in our daily lives, spanning across diverse industries as well as academia. In the field of high energy physics (HEP), the most common and challenging task is separating a rare signal from a much larger background. The boosted decision tree (BDT) algorithm has been a cornerstone of the high energy physics for analyzing event triggering, particle identification, jet tagging, object reconstruction, event classification, and other related tasks for quite some time. This article presents a comprehensive overview of research conducted by both HEP experimental and phenomenological groups that utilize decision tree algorithms in the context of the Standard Model and Supersymmetry (SUSY) . We also summarize the basic concept of machine learning and decision tree algorithm along with the working principle of Random Forest, AdaBoost and two gradient boosting frameworks, such as XGBoost, and LightGBM. Using a case study of electroweakino productions at the high luminosity LHC, we demonstrate how these algorithms lead to improvement in the search sensitivity compared to traditional cutbased methods in both compressed and non-compressed R-parity conserving SUSY scenarios. The effect of different hyperparameters and their optimization, feature importance study using SHapley values are also discussed in detail. Contents 1 Introduction 1 2 Basic concepts of machine learning 3 2.1 Loss Function 4 2.2 Overfitting and underfitting 5 2.3 Measures of classification performance 6 3 Machine learning in High Energy Physics 9 3.1 Signal and background events classification 12 3.1.1 Searches for RPC SUSY scenarios using BDT 13 3.1.2 Searches for RPV SUSY scenarios using BDT 19 4 Decision Tree algorithms 20 4.1 Random Forest 23 4.2 AdaBoost 25 4.3 XGBoost 27 4.4 LightGBM 29 5 Performance of different Decision Tree based algorithms-a RPCSUSY case study at the HL-LHC 31 5.1 Cut-and-count analysis 34 5.2 Machine Learning based analysis 36 5.2.1 Hyperparameter variation for different algorithms 37 5.2.2 Feature importance with SHapley 39 5.2.3 Comparison of results coming from different algorithms 41 6 Summary 43 Bibliography 45 7 Appendix 58 1 Introduction Supersymmetry (SUSY) [1–3] is one of the most compelling extensions of beyond the Standard Model (BSM) scenario, and the pursuit of supersymmetric partners of the SM particles (sparticles) remains as a primary objective at the Large Hadron Collider (LHC) . Since the inception of the LHC, both the ATLAS and CMS collaborationshave already conducted numerous searches to explore the SUSY particles utilizing the LHC Run-I and Run-II dataset [4 , 5] . In the absence of any statistical deviations from the SM predictions, the LHC has set stringent lower bounds on the masses of particles. For example, in the R-parity conserving (RPC) SUSY 1 scenarios with relonstoTeattpVihrvee(meslymet1pae)ligssacthtesndiveneofthlyue[tgll4rauig, 5ilnh]inootesdes(mtpe(c01)ad,)ri,fignthrsingetooLtn(Hwomth1gRunener) uptbran-IIatochal3raihas eght sTeVtios,xtendedquarks1 6 TeVand thet(s,hm1imeqe2p)l,oTliwertheeVfied blamiognuhdondtes1delst2s assumptions. It is important to note that the accumulated luminosity of Run-II data is approximately ∼ 140 fb−1, which is about 5% of the luminosity of planned upgrade of high luminosity LHC (HL-LHC) run (L = 3000fb−1) . The LHC Collaboratio

FAQs

Why are decision tree based machine learning methods important for BSM searches at the LHC?

They help separate rare signal events from much larger backgrounds, improving classification performance. The paper highlights their role in BDT-based and gradient-boosting searches for SUSY scenarios.

Which decision tree frameworks are compared in the study?

Random Forest, AdaBoost, XGBoost, and LightGBM are reviewed and compared. Their principles and how they are used in HEP classification are summarized.

How does the case study on electroweakino production at the HL-LHC evaluate performance?

It demonstrates improved search sensitivity compared with traditional cut-and-count methods in both compressed and non-compressed R-parity conserving SUSY scenarios. It also analyzes hyperparameter effects and feature importance using SHapley values.

Aug 5, 2026

Upload by Asher