Machine learning predictions of cross-sections for the ²³²Th(p,x)²²⁵Ac reaction

The research investigates techniques to enhance actinium-225 (²²⁵Ac) production levels for targeted alpha therapy (TAT) applications in cancer medicine. The medical benefits of ²²⁵Ac alpha emissions and favorable half-life exist despite production limitations which restrict its clinical applications. Proton-induced spallation of thorium-232 (²³²Th) shows promise as a commercially viable method but scientists require improved ways to predict cross-section data accurately. The researchers refined TALYS-2.0 code simulations by making manual parameter changes to match experimental data results more closely. Cross-section experimental data from curated sources served to develop four various machine learning (ML) models including support vector regression (SVR), random forest (RF), artificial neural network (ANN) and polynomial regression (PR). The reaction cross-sections and physical yields of ²²⁵Ac showed best prediction results through the usage of ANN and SVR. The investigation evaluated both the manufacturing outcome and purities while specifically measuring the ²²⁷Ac/²²⁵Ac proportion because this factor determines medical usefulness. The research findings demonstrate that ML applications increase predictive models to develop practical methods that optimize ²²⁵Ac production while lowering impurities which improves its clinical suitability for TAT treatment.