Development of a model for design radiation shielding composite aprons using machine learning

This study presents the development and evaluation of a novel lead-free composite for radiation shielding, designed using an artificial neural network (ANN). The ANN model achieved a high predictive accuracy for the mass attenuation coefficient, with a correlation coefficient of R = 0.997 and a mean absolute percentage error of 0.9-1.8%, closely matching benchmark Geant4 Monte Carlo simulations. The optimized Ba-Gd-I-Sb composite (73.9% Ba, 18.2% Gd, 7.65% I) exhibits superior shielding by exploiting synergistic multi-edge absorption. Key quantitative results show that a 1-mm thickness provides > 10⁵ attenuation at 50 keV, > 95% absorption at 1.0 mm thickness, and a favorable weight-to-performance ratio (1.08 kg for a 0.5 mm shield). The composite's attenuation of 99% at 80 keV exceeds that of commercial non-lead alternatives (97-98%), while its transmission matches that of lead at lower energies. These findings demonstrate the composite's high performance and potential as a sustainable, efficient solution for medical and industrial radiation protection.