Experimental data-driven modeling and prediction of (γ,n) cross-sections with physics-informed neural networks and gradient boosted decision trees

In this study, we present a hybrid machine learning framework for modeling and predicting (γ,n) reaction cross sections across the nuclear chart, combining physical interpretability with predictive power. The first stage uses a Physics-Informed Neural Network (PINN) to fit experimental data from EXFOR while enforcing domain-specific constraints, such as energy thresholds, Lorentzian resonance behavior, and monotonicity beyond the Giant Dipole Resonance (GDR). These high-quality fits are then used to train a Gradient Boosted Decision Tree (GBDT) model on a broad set of nuclear features including mass, shell effects, separation energies, and deformation parameters. The resulting model agrees well with both experimental and evaluated data for known isotopes and extrapolates plausibly to exotic nuclides lacking measurements. Case studies on stable isotopes, actinides, and neutron-rich nuclei demonstrate the model’s robustness. This approach illustrates the complementarity of physics-informed and data-driven modeling for improving cross-section coverage in nuclear physics, simulation, and security applications.