Prediction of mass attenuation coefficients in mixed alkali and borosilicate glasses using machine learning approaches

Abstract

The increasing demand for effective radiation shielding materials in different sectors, including medical, nuclear, and industrial applications, has promoted exploring new approaches for optimizing material properties. Mixed alkali glasses, such as those containing B2O3, SiO2, CaO and ZnO, offer a combination of protective and transparent properties, making them promising candidates for radiation shielding. Recent advances in machine learning (ML) have accelerated the development and evaluation of these materials. This study utilizes ML techniques to predict the mass attenuation coefficients (MAC) of mixed alkali glasses across a photon energy range of 0.015 MeV–15 MeV. A comprehensive dataset generated using XCOM served as the basis for training and validating several ML models, including Linear Regression, Ridge Regression, Lasso Regression, Support Vector Regression, Random Forest Regression, K-Nearest Neighbors, Multi-Layer Perceptron, and a customized Neural Network (NN). Among these models, Random Forest Regression ended as the most accurate, achieving an R-squared value of 0.99958 and demonstrating minimal error (MAE: 0.01481, MSE: 0.00561, RMSE: 0.07493), indicating its superior ability to capture the complex relationships between glass composition, energy levels, and radiation attenuation properties. While other models like MLP and NN performed decently, they lagged behind the Random Forest model. The results highlight machine learning's potential to advance radiation shielding by providing reliable models for material design and parameter calculations like MAC, offering faster, more efficient alternatives to conventional tools like Monte Carlo simulations.