Machine learning-driven identification of E1 moments from photonuclear cross-section measurements

Abstract Machine learning (ML) has become a new paradigm in scientific research across various fields, with recent nuclear structure physics studies increasingly focusing on its powerful ability to effectively model large and small datasets. This study aims to predict the electric dipole (E1) energy-weighted photonuclear cross-section moments (σ 0 , σ -1 , σ -2 ) and E1 polarizabilities (α E1 ) in the actinide nuclei using Artificial Neural Networks (ANN) with Levenberg-Marquardt (LM), Scaled Conjugate Gradient (SCG), and Bayesian Regularization (BR) algorithms, as well as Adaptive Neuro-Fuzzy Inference System (ANFIS) methods. The dataset, derived from photonuclear cross-section libraries and last updated in 2019, features various nuclei and includes input parameters that describe their characteristics. The data are partitioned into 70% for training, 15% for validation, and 15% for testing, with robust model evaluation. The performance of the ANN and ANFIS models is assessed using relevant metrics. Furthermore, we examine potential applications of ML in predicting E1 photonuclear cross-section moments and offer predictions for certain odd-A deformed actinide nuclei that lack experimental data, using the most effective models from various ML algorithms. The predictive results obtained from ML are critically compared with the outcomes of the Translational and Galilean Invariant-Quasiparticle Nuclear Model (TGI-QPNM), highlighting the efficacy and potential of ML methodologies in advancing nuclear physics research.