Machine learning-enhanced nuclear charge radius predictions: A comprehensive comparative analysis

In this study, we systematically apply various machine learning (ML) algorithms to enhance the calculation of nuclear charge radii. A comprehensive comparison within a broad spectrum of ML methods, including support vector regressor, k-nearest neighbors, decision trees, random forest, gradient boosting regressor and extreme gradient boosting machine, is conducted to evaluate their predictive performance. A grid search approach is also performed to identify the best hyperparameter combination for each ML algorithm. Our results indicate that methods based on gradient boosting are particularly advantageous for learning the residuals, i.e., deviations between experimental and calculated charge radii. Since the charge radius exhibits intricate dependencies on both proton and neutron numbers, gradient boosting methods, by sequentially refining weak learners, can effectively approximate these subtle dependencies, making them well-suited for improving charge radius calculations across different nuclei.