Hybrid machine learning-monte carlo approach for neutron shielding in high-energy proton facilities

Journal of Subatomic Particles and Cosmology, cilt.5, sa.2026, ss.1-14, 2026 (Hakemli Dergi)

target materials constitutes a critical challenge for radiation safety and shielding design. In this study, a hybrid

approach combining Monte Carlo (MC) simulations with machine learning (ML) techniques was developed to

enable rapid and reliable prediction of neutron dose distributions. Using the FLUKA code, a total of 598 datasets

were generated for 250 MeV and 1000 MeV proton energies, covering iron (Fe) and ferroboron (Fe₂B) shielding

configurations. These datasets were employed for the training and validation of Logarithmic Linear Regression

(log-linear LR), Gradient Boosting Regressor (GBR), K-Nearest Neighbors (KNN), and Random Forest (RF)

models. The RF and GBR algorithms demonstrated high predictive accuracy (R² ≈ 0.95–0.99) while significantly

reducing computational time compared to conventional MC simulations. The results indicate that ML-assisted

approaches provide a reliable, cost-effective, and time-efficient alternative for neutron shielding design in nuclear

facilities.