Machine-learning predictions of fusion cross sections for synthesizing 99−103Mo*

This dataset is designed to support research on the application of machine learning methods in predicting fusion cross-sections for medium-low energy heavy-ion reactions, with a focus on fusion reactions related to 99–103Mo* and the calculation of neutron evaporation channels. The dataset encompasses experimental data, results from theoretical model calculations, and predictions from machine learning models, which can be used for model training, comparative validation, and reproduction of subsequent methods. The data generation period spans from 2024 to 2025, during which both theoretical model calculations and machine learning predictions were completed.The construction process of this dataset is as follows: First, fusion reaction experimental cross-section data and basic physical parameters of related nuclides are compiled from public nuclear data resources and published literature; second, the phenomenological theoretical model (EBD2) is used to calculate the fusion cross-sections for corresponding reactions at different energy points; on this basis, a LightGBM machine learning model based on gradient boosting decision trees is employed to perform batch predictions of fusion cross-sections for 99–103Mo* reactions; finally, combined with the neutron evaporation channel survival rates calculated by the Gemini++ program, the evaporation residue cross-section data for nuclides such as 99Mo are obtained. All data files are stored in CSV or Excel format and can be opened directly using common data analysis software.File One: Synthesis of Mo Isotopes Projectile and Target Nuclei Half-Life Data Table.csv. This file contains basic nuclide information for target and projectile nuclei related to the synthesis of Mo isotopes. Data field descriptions are as follows: Z: proton number of the nuclide; A: mass number of the nuclide; N: neutron number of the nuclide; T: half-life of the corresponding nuclide.Data sourced from the public nuclear database NRV (Nuclear Reaction Video / Nuclear Data Resource). The data acquisition method is: search for corresponding entries in the NRV database based on nuclide names, and manually verify, organize, and enter them. This file is mainly used to provide basic nuclide parameters for fusion reaction and evaporation process analysis.File Two: Partial Fusion Reaction Experimental Data and Theoretical Data Table.csv. This file summarizes cross-section data for six typical fusion reaction systems, including: 12C + 92Zr, 27Al + 72Ge, 27Al + 73Ge, 16O + 74Ge, 35Cl + 54Cr, 40Ca + 50Ti. The file simultaneously includes data from the following three sources: experimental measured cross-section data from published literature, cross-section data calculated by the EBD2 phenomenological theoretical model, and cross-section data predicted by the machine learning model. Main field descriptions are as follows: z: proton number of the incident nucleus; a: mass number of the incident nucleus; MeV: center-of-mass energy of the reaction.This file is mainly used for horizontal comparative analysis and model validation among different methods.File Three: Batch Prediction of 99Mo Fusion Reaction Cross-Section Data.csv. This file is used to display batch prediction results for 99Mo fusion reaction cross-sections, with data sources including both theoretical models and machine learning models. The file contains two worksheets (Sheets):Sheet1: Fusion reaction cross-section data generated by the EBD2 theoretical model online calculation tool (calculation URL: http://www.imqmd.com/fusion/EBD2.html, data acquisition up to December 2025); Sheet2: Fusion reaction cross-section data predicted by the LightGBM machine learning model at the same energy points. Main field descriptions are as follows: Z: proton number of the nuclide; A: mass number of the nuclide; E column: center-of-mass energy of the reaction; F column: fusion reaction cross-section at the corresponding energy point.This file can be used to compare the prediction differences between theoretical models and machine learning methods under the same physical conditions.File Four: Machine Learning Prediction of 100–103Mo Fusion Reaction Cross-Sections, Gemini++ Neutron Reaction Channel Survival Rates, 99Mo Evaporation Residue Cross-Sections.csv. This file contains multiple worksheets, systematically displaying the complete data chain from fusion cross-section prediction to evaporation residue cross-section calculation, as follows:The first four Sheets: Correspond respectively to machine learning prediction results for fusion reaction cross-sections in synthesizing 100Mo, 101Mo, 102Mo, and 103Mo.Z: proton number of the nuclide; A: mass number of the nuclide; E column: reaction energy (MeV); F column: fusion reaction cross-section at the corresponding energy point;The fifth to eighth Sheets: Correspond to survival rate data for 1n–4n neutron evaporation channels, with data calculated by the Gemini++ program. A column: energy point; B: survival rate for the corresponding neutron evaporation channel;The last four Sheets: Evaporation residue cross-section data obtained by multiplying the fusion reaction cross-sections with the corresponding evaporation channel survival rates. Z: proton number of the nuclide; A: mass number of the nuclide; E column: center-of-mass energy of the reaction; F column: evaporation residue cross-section at the corresponding energy point.This file provides complete data support for researching reaction path selection and cross-section estimation for the medical isotope 99Mo.On the test set, the mean absolute error (MAE) between the machine learning model's predicted fusion reaction cross-sections (CS) and experimental values is 0.0615, which is superior to the EBD2 model's 0.1103. This article demonstrates the superiority of machine learning in predicting nuclear fusion reactions, providing a reliable tool for the optimal synthesis paths of the medical isotope 99Mo. In the future, the model can be extended to more nuclides, integrate real-time experimental data, improve prediction accuracy, and explore integration with other theoretical models to support practical applications in nuclear medicine.