Implementation of machine learning models for prediction of primary gamma-ray emissions from thermal neutron capture reactions.

The National Nuclear Data Center (NNDC) at Brookhaven National Laboratory maintains a dedicated effort towards providing easy access of reliable and evaluated nuclear data. Nuclear data includes features describing the lifetime, mass, types of decays, and energies related to a nuclide. Optimizing fuel, space exploration, and nuclear nonproliferation depend upon a precise capture gamma-ray evaluated library. Using the Evaluated Gamma-ray Activation File (EGAF) and NuDat, the goal of this project is to employ machine learning methods to predict primary gamma ray emissions from isotopes which have undergone neutron capture at thermal incident energies. A neutron capture reaction is identified by the formation of a residual nucleus composed by the projectile neutron and the target isotope. This new system is presumed to be formed at an excited state of the compound nucleus and undergoes a sequence of gamma ray emissions as a mean of de-excitation, until eventually reaching its ground state. To comprehend the cascade of emissions, it is most significant to understand the first steps in the decay­: the primary emissions. Multiple characteristics of an isotope, relevant to its susceptibility to neutron capture or lack thereof, can greatly enhance our comprehension of nuclei and their inherent properties. Instead of using standard theoretical nuclear models, training classification and regression models using machine learning algorithms –such as Random Forest and K-Nearest Neighbors– with experimental data can allow for the machine to predict primary gamma ray emission probabilities missing in data sets with greater accuracy. Upon successful project completion the NNDC will be able to provide more precise capture gamma-ray spectra in evaluated libraries. Throughout this project, I became skilled in data analysis, mining, and visualization in addition to implementing and optimizing different machine learning models in Python.