@article{article_1945462, title={Development and Validation of a Random Forest Model for X-Ray Shielding Thickness Prediction in Myanmar Healthcare Facilities Based on NCRP Report 147}, journal={Journal of Nuclear Sciences}, volume={10}, year={2026}, DOI={10.59474/nuclear.2023.69}, url={https://izlik.org/JA26NJ24UP}, author={Oo, Zaw Lin and Laı, Theint Win}, keywords={X-ray shielding, Radiation protection, Diagnostic radiology, NCRP Report 147, Machine learning.}, abstract={<p>This study aims to develop and validate a machine learning model for accurate and rapid prediction of required brick wall thickness for diagnostic X-ray rooms, based on the methodology of NCRP Report No. 147, and to assess its performance using real-world clinical data from Myanmar. A synthetic dataset of 864,000 samples was generated based on the shielding design methodology in NCRP Report No. 147. A Random Forest regressor was trained and hyperparameter-tuned, then benchmarked against baseline models including Linear Regression and Decision Trees. The final model was externally validated using real-world data from 11 diagnostic X-ray installations (4 polyclinics, 4 general hospitals, and 3 diagnostic centers) across Myanmar. The best-performing model was deployed as a free open-access web application on Hugging Face Spaces. The optimized Random Forest model achieved excellent performance on the synthetic test set (MAE = 0.056 cm, RMSE = 0.156 cm, R² = 0.9996), markedly outperforming the best baseline model (Decision Tree, MAE = 0.615 cm). Feature importance analysis showed that tube voltage (kVp) and brick density were the dominant predictors, contributing 53.2% and 37.8% of the total importance, respectively. External validation on independent real clinical data yielded a mean absolute error of 3.22 cm, which remains within conservative safety margins for shielding design. The model also indicated that conventional 9-inch (22.9 cm) brick walls are frequently inadequate for typical clinical workloads in Myanmar. This study demonstrates that a well-trained Random Forest model can provide highly accurate and rapid predictions of X-ray shielding requirements. Through successful validation with real clinical data from a low-resource setting and the provision of an open-source web application, the proposed approach offers a practical and accessible tool for radiation safety officers and facility designers in Myanmar and similar healthcare environments. </p>}, number={2}, organization={Department of Atomic Energy, Myanmar and Ascend International Preparatory College,}