Nuclear Mass Predictions through Neural Networks Incorporating Neutron and Proton Separation Energy Constraints

This dataset is generated by combining nuclear mass theoretical models with machine-learning techniques. The data production process consists of two main steps: (1) constructing training samples based on existing nuclear mass databases (such as AME2020 and PCF-PK1); and (2) training and predicting with four independently built artificial neural network models (ANN1–ANN4).The training samples cover a range of nuclides determined by the proton number Z and neutron number N, spanning from light nuclei to the superheavy region. Since the ANN1–ANN4 networks are trained independently, the dataset provides four separate sets of prediction results, which can be used to reflect uncertainties arising from model-structure variations.Data FilesThe dataset contains four primary data files, corresponding to the prediction results of the four neural network models:ANN1_extra.txtANN2_extra.txtANN3_extra.txtANN4_extra.txtAll files are in TXT format (comma-separated text) and can be opened with any text editor, Microsoft Excel, or scientific computing tools such as Python/pandas or MATLAB, without requiring any special software.Each file contains the following columns:Z, N: proton number and neutron number of the nucleus;E: binding energy;Sn, S2n: one-neutron and two-neutron separation energies;Sp, S2p: one-proton and two-proton separation energies.All binding energies and separation energies are given in MeV.