Enhancing precision in J/ψ mass estimation: A study of ensemble and deep learning methods

Abstract

This study evaluates ensemble learning methods and Deep Neural Networks (DNNs) for identifying J/$\psi \to {\mu }^{+}{\mu }^{-}$ events in proton-proton collisions at the LHC, focusing on the dimuon decay channel within a skewed dataset. For this purpose, 8 different machine learning models based on Random Forest (RF), Gradient Boosting Decision Trees (GBDT), and DNNs were implemented to investigate the most effective approach for charmonium event determination. Performance metrics such as precision, recall, F-1 Score, geometric mean (G-mean), and balanced accuracy (BAcc) are employed, with StratifiedKFold cross-validation verifying the models' robustness in skewed data scenarios. Results demonstrate DNNs as the most proficient, underscoring their potential in complex data analysis in particle physics. Utilizing the Crystal Ball (CB) function on the results of DNNs, the precision of the J/ψ mass was estimated. This study not only enhances understanding of machine learning applications in high-energy particle collisions but also sets the stage for more advanced research in this field.