Predictive Modeling of Radiation Shielding Properties for Lanthanum-Cerium Oxide Doped 45S5 Bioactive Glass Using Machine Learning Algorithms

This study investigates the incorporation of cerium oxide (CeO2) and lanthanum oxide (La2O3) into 45S5 bioactive glass to enhance its radiation shielding properties, critical for nuclear and biomedical applications. Traditional evaluation methods, including simulations and experimental testing, are often resource-intensive. To address this, a machine learning approach was employed to predict these properties in (CeO2+La2O3)-doped 45S5 bioactive glass. Several Machine learning models, including Random Forest, LightGBM, Support Vector Regressor, CatBoost, and XGBoost, were trained and validated using data from EpiXS and Monte Carlo Geant4 simulations, achieving high predictive accuracy. The machine learning model results, including predictions for mass attenuation coefficients (MAC), which serve as key indicators of radiation shielding effectiveness, were compared to Monte Carlo Geant4 simulations and the EpiXS database, demonstrating strong agreement. Among the evaluated models, CatBoost, Extra Trees, LightGBM, and Random Forest demonstrated solid predictive accuracy, with R2 scores of 0.9999, 0.9998, 0.9996, and 0.9996 , respectively. Although slight deviations were observed at very low photon energies (<0.1 MeV), reflecting the increased complexity of photon-matter interactions in this region. These results highlight the potential of machine learning to improve the optimization of doped bioactive glasses, significantly reducing the time and cost of conventional methods. This approach enables the rapid development of bioactive glass compositions with enhanced mechanical and radiation shielding properties, advancing biomedical implants and protective materials for radiological applications.