Precise values of binding energies play an important role in computing different astrophysical processes, which improves our understanding of the universe. There are a limited number of nuclei that have been experimentally determined. Different models can predict the binding energies of any nuclei. However, making models that predict binding energies with a high degree of accuracy without a large computational expense is a difficult task. This thesis will present different models and methods used to improve predictions for nuclear binding energies. Our overarching approach was to take a simple model and add complexity to increase its accuracy. We started with the liquid drop model and the accompanying semi-empirical mass formula as it is a simple function of the number of protons and neutrons and still leads to fairly accurate results. We used machine learning techniques with different regression techniques, such as linear regression, ridge regression, and support vector regression to model the errors between the experimental nuclear binding energy data and those predicted by the liquid drop model. This error model is then tacked onto the semi-empirical mass formula to increase its predictive accuracy. We saw that support vector regression with Gaussian features centered around magic numbers produced the best error model as it was able to take into account nuclear shell structure not accounted for in the liquid drop model.
