Prediction of binding energy using machine learning approach

Abstract

The liquid drop model is an empirical hypothesis established on the idea that nuclei can be thought of as incompressible liquid droplets. The AME2020 dataset was used in this work to determine binding energy using a semi-empirical mass formula and compare it with binding energies predicted by a machine learning algorithm. Random forest regressor, MLPRegressor, and XGBoost models were employed. In terms of accuracy, root mean square error, and mean absolute error, machine learning models performed better than the semi-empirical mass formula. Compared to RFR, XGBoost, and SEMF, MLPRegressor performed better in predicting binding energies for lighter nuclei. Using estimated binding energies, nuclear masses were computed, and it was shown that all three models adequately predicted nuclear masses with minimal error. This finding highlights how machine learning can be applied to nuclear physics to predict various nuclei’s properties.