Predicting hadron mass and decay width using XGBoost: A unified machine learning framework for baryons and mesons
Abstract
This study presents an XGBoost-based machine learning framework for predicting the mass and decay width of hadrons, including both baryons and mesons, within a unified feature space. The dataset was compiled from Particle Data Group (PDG) records and contains 495 hadrons in total, reduced to 406 and 373 usable samples for mass and width prediction tasks respectively after data cleaning. Model inputs consist of quark and antiquark counts along with conserved quantum numbers including isospin, spin, parity, electric charge, strangeness, charmness, and bottomness. Analyses were conducted at three levels: baryon-only, meson-only, and combined hadron datasets. For width prediction, two scenarios were evaluated: one excluding and one including hadron mass as an input feature. Mass prediction achieved high accuracy across all datasets, with test R² values of 0.904, 0.959, and 0.964 for baryons, mesons, and the combined hadron dataset, respectively. Feature importance analysis identified heavy-flavor quark content, particularly bottom quarks, as the dominant factor in mass prediction. Width prediction without mass information yielded limited explanatory power, especially for mesons (R²=0.203), whereas including mass as an input substantially improved performance, raising the combined hadron model's test R² to 0.812. These results confirm that hadron mass plays a central role in governing decay width through kinematic decay channels. Data augmentation via Gaussian-noise perturbation did not improve generalization over the baseline model trained on real observations alone. The study demonstrates that XGBoost offers both predictive accuracy and physical interpretability, providing a complementary data-driven tool for hadron spectroscopy.
Keywords
Hadron spectroscopy, XGBoost, mass prediction, decay width, machine learning
References
- Akan, T. (2024). Exploring machine learning models for predicting meson mass and width. Physica Scripta, 99(2), 025203. https://doi.org/10.1088/1402-4896/ad1902
- Chen, T., & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 785–794). ACM. https://doi.org/10.1145/2939672.2939785
- Malekhosseini, M., Rostami, S., Olamaei, A. R., Ostovar, R., & Azizi, K. (2024). Meson mass and width: Deep learning approach. Physical Review D, 110, 054011. https://doi.org/10.1103/PhysRevD.110.054011
- Navas, S., Amsler, C., Gutsche, T., Hanhart, C., Hernández-Rey, J. J., Lourenço, C., Masoni, A., Mikhasenko, M., Mitchell, R. E., Patrignani, C., et al. (Particle Data Group). (2024). Review of particle physics. Physical Review D, 110, 030001. https://pdg.lbl.gov/
- Nielsen, D. (2016). Tree boosting with XGBoost: Why does XGBoost win "every" machine learning competition? [Master's thesis, Norwegian University of Science and Technology]. NTNU Open. https://ntnuopen.ntnu.no/ntnu-xmlui/handle/11250/2433761
- Particle Data Group. (2020). 2020 review of particle physics: Interactive listings. https://pdg.web.cern.ch/pdg/2020/listings/contents_listings.html
- Tong, X., Feng, W., Xu, W., & Li, X. Q. (2025). Meson properties and symmetry emergence based on the deep neural network. Physical Review D, 111, 014023. https://doi.org/10.1103/PhysRevD.111.014023
- Workman, R. L., Burkert, V. D., Crede, V., Klempt, E., Thoma, U., Tiator, L., Agashe, K., Aielli, G., Allanach, B. C., Amsler, C., et al. (Particle Data Group). (2022). Review of particle physics. Progress of Theoretical and Experimental Physics, 2022, 083C01. https://doi.org/10.1093/ptep/ptac097
- Yasser, A. M., Nahool, T. A., Anwar, M., & Mansour, N. A. (2020). A new machine learning approach for predicting the spectra of meson bound states. International Journal of Modern Physics A, 35(32), 2050214. https://doi.org/10.1142/S0217751X2050214X