Using Deep Learning to Predict the Half-Lives of Isotopes Given Proton and Neutron Count

This study explores using deep learning regression models to predict superheavy isotopes' half-lives based on proton and neutron counts, finding that additional complexity in input variables and model architecture may be necessary to improve prediction accuracy.

In this study, we attempt to predict the half-lives of superheavy isotopes through regression models. Superheavy elements usually have very short half-lives, but a few isotopes are predicted to have long enough half-lives for potential applications in nuclear and material science. To calculate these half-lives, we initially used a deep learning regression model with Keras library’s dense layers that took proton count and neutron count as inputs and $\log_{10}$ of the half-life as the output. When it was tested on various isotopes of the periodic table, there was an MSE (mean squared error) of 111. Afterwards, more input variables related to proton and neutron count were added to increase complexity, leading to an MSE of 97. This suggests that the model on average falls many orders of magnitude off from the actual half-life, which would make accurate prediction of half-lives difficult. It is possible that more complexity, through different kinds of layers, and different inputs may be needed to accurately predict the half-lives of nuclides (another word for isotopes).

Explore More!

Research Paper