Machine Learning Prediction of 52Cr and 183W (n,2n) Reaction Cross Sections Using Experimental and Evaluated Nuclear Data
Year: 2026
Type: book
Abstract: Accurate neutron-induced reaction cross sections are essential for nuclear-data assessment and a range of applications involving neutron interactions. In this study, five machine-learning regression algorithms—Bayesian Regularized Neural Network (BRNN), Cubist, Random Forest (RF), Support Vector Regression (SVR), and eXtreme Gradient Boosting (XGBoost)—were evaluated for predicting the excitation functions of the 52Cr(n,2n)51Cr and 183W(n,2n)182W reactions. Models were trained using combined experimental and evaluated nuclear data compiled from the EXFOR database and major nuclear-data libraries and assessed using held-out test data through deviation analysis, residual distributions, predicted-versus-observed agreement, and complementary regression metrics. Across both reactions, Cubist and BRNN consistently provided the best predictive performance with RMSE values below 74 mb and r2 ≥ 97%, followed by RF and XGBoost, whereas SVR showed substantially larger errors and systematic deviations, particularly at higher cross-section values. For the 52Cr and 183W datasets, Cubist achieved MAE values of 22.9 and 52.1 mb, respectively, while SVR yielded MAE values of 93.9 and 274.9 mb. The machine-learning predictions were additionally compared with TALYS calculations, providing a physics-based consistency benchmark for the overall excitation-function trends. The results indicate that appropriately selected machine-learning models can reproduce important features of (n,2n) excitation functions across substantially different cross-section ranges and may provide rapid, computationally inexpensive support for cross-section interpolation, preliminary estimation in sparsely measured energy regions, and nuclear-data assessment. These models are intended as complementary surrogate or screening tools rather than replacements for evaluated nuclear-data libraries or physics-based nuclear reaction codes.
Institutions: Giresun University, Necmettin Erbakan University
Language: English
Subfield: Nuclear and High Energy Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204893145