In collider experiments, the production of J/$\psi$, a bound state of charm and anti-charm quarks (c\$bar{c}$), can serve as the testing ground for the theory of strong interaction. In experiments, J/$\psi$ meson production yield is preferably estimated by reconstructing it from its electromagnetic decays to dileptons, i.e., J/$\psi\rightarrow\mu^++\mu^-$ or J/$\psi\rightarrow e^++e^-$. This inclusive production of J/$\psi$ includes both prompt and non-prompt contributions. The prompt J/$\psi$ are produced in the initial hadronic collisions or via feed-down from directly produced higher charmonium states. In contrast, the non-prompt J/$\psi$ are the products of weak decays of the beauty hadrons. This study uses machine learning (ML) based models such as the XGBoost and LightGBM to identify the inclusive, prompt and non-prompt production of J/$\psi$ from the uncorrelated background dimuon pairs. The model is trained using PYTHIA8 generated data for proton-proton (pp) collisions at $\sqrt{s}=13$ TeV. The inputs to both the ML models include the observables that are easily obtained in the experiments. Both models attain a prediction accuracy of up to 99% while retaining the transverse momentum, pseudorapidity and collision energy dependence. In addition, the models can be applied to identify each dimuon pair separately having any value of transverse momentum in any rapidity range. Thus, it can probe the production fraction of non-prompt J/$\psi$ ($f_B$) in smaller $p_T$ and rapidity bins. In addition, this dimuon level tagging can help us study many physics aspects, such as polarization and flow measurements of prompt and nonprompt J/$\psi$ with ease experimentally, which is an advantage of using this method.
Reference: S. Prasad, N. Mallick and R. Sahoo, Phys. Rev. D 109, 014005 (2024).