Integrating physics insights into machine learning: a case study with giant dipole resonance

Abstract Utilizing a recent compilation of experimental data, selected simple theoretical models, and machine learning (ML) algorithms, we analyze a few ways to construct a model with better accuracy and predictive power. For the data and theoretical models, we consider the giant dipole resonance (GDR), a predominant feature of photo-nuclear interactions. We study the dependence of the dipole-dipole interaction strength on the mass number and the energy dependence of the GDR width. The best ML algorithm is chosen among the considered ones by comparing different error metrics. Employing GDR parameters obtained using the best algorithm, we also calculate the cross-section and compare it with the data. For validating our results, we choose even–even Nd isotope chain 144−152 Nd and 152 Sm. We propose that a model constructed with appropriate physics inputs where the parameters are optimized through ML is successful in explaining the data and has better predictive power.