Nuclear mass excesses calibrated by Bayesian ridge regression for predicting optimal incident energies for the synthesis of superheavy elements

We develop a machine learning method to calibrate nuclear mass excess predictions by integrating multiple global mass models. A Bayesian ridge regression model is trained with known nuclei from AME2020 using results from five theoretical mass models, achieving a root-mean-squared error of 0.256 MeV. Further validations based on updated mass evaluations, excluded nuclear chains, and extrapolation toward the superheavy region support the improved interpolation of the Bayesian ridge regression model among known nuclei and improved extrapolation performance. The calibrated results are then introduced into the dinuclear system model to enhance the prediction of optimal incident energies. This machine learning--assisted approach provides reliable optimal incident energies for the synthesis of superheavy elements with $Z=119$ and 120. Our study indicates the potential of combining theoretical models with machine learning methods to advance predictive capabilities in superheavy element researc