Data selection tools based on machine learning for hyperon form factor studies with the Belle II experiment

Sumitted to PubDB: 2025-05-08

Category: Master Thesis

Tags:

Principal Authors	Federico Bonaldo
Non-Belle II authors	Federico Bonaldo
Date	2024-04-22
Belle II Number	BELLE2-MTHESIS-2025-013
Abstract	Study the internal structure of the Λ hyperon at the energy of Belle II may be possible through the channel e+e− → Λ¯ΛγISR. The reconstruction for this channel is fundamental for the scope of the project. Hence, it is crucial to investigate the reconstruction of the γISR because including its information on the final state may improve the reconstruction analysis for this channel. Whereas hyperons are unstable particles, we use a reconstruction algorithm to access the Λ particles. Since the reconstruction algorithm collects a lot of background caused by a wrong reconstruction, in my thesis I developed a Λ hyperon selector using machine learning tools aimed to improve the current reconstruction algorithm for the Λ hyperons. The analysis I conducted uses a supervised learning model, trained and tested on Monte Carlo data samples, to study which features, i.e. kinematics variables, are relevant for the selector. Thanks to the Monte Carlo Truth used as a target variable, I evaluated the performance of my Λ selector reaching an efficiency of 80% and a purity of 90% on an independent Monte Carlo sample. The work of my thesis proves the great potential of the application of machine learning tools in high energy physics experiment.