Support vector regression modeling of (n,t) reaction cross sections at 14–15 MeV using physically informed descriptors
Applied Radiation and Isotopes, vol.237, 2026 (SCI-Expanded, Scopus)
- Publication Type: Article / Article
- Volume: 237
- Publication Date: 2026
- Doi Number: 10.1016/j.apradiso.2026.112869
- Journal Name: Applied Radiation and Isotopes
- Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Chemical Abstracts Core, Chimica, Compendex, EMBASE, INSPEC, MEDLINE, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO)
- Keywords: (n,t) reaction cross section, Machine learning, Semi-empirical cross section formula, SHAP analysis, Support vector regression
- Akdeniz University Affiliated: Yes
Abstract
In this study, (n,t) reaction cross sections induced by 14–14.8 MeV neutrons were modeled using the Support Vector Regression (SVR) method with physically meaningful nuclear and reaction parameters. The input features of the model include neutron number (N), proton number (Z), mass number (A), incident neutron energy (En), threshold energy (Eth), reaction Q-value, asymmetry parameter S = (N−Z)/A, and Coulomb energy (EC). The SVR approach achieved a high coefficient of determination of R2 = 0.9868, with a Mean Absolute Error (MAE) of 54.65 μb and a Root Mean Square Error (RMSE) of 171.62 μb, indicating stable predictive performance for the analyzed dataset. Comparative analysis shows that the SVR model provides a better agreement with experimental data compared to the selected semi-empirical formulas. Furthermore, SHapley Additive exPlanations (SHAP) analysis was used to interpret the feature sensitivity of the trained model, identifying Coulomb energy (EC) and reaction Q-value as the most influential descriptors. The results demonstrate that a data-driven SVR approach based on physically grounded input parameters can provide balanced and interpretable predictions compared to fixed-parameter semi-empirical models.