Nuclear $\beta$-decay energies $Q_{\beta }$ are investigated using the machine learning algorithm LightGBM. By comparing different combinations of physically informed input features, the optimal feature set for predicting $Q_{\beta }$ was obtained. The mean absolute error reached 0.1447 MeV and demonstrated strong interpolation capability in the region of $A>50$. To address the black-box problem, feature analysis is performed using the interpretable Shapley Additive Explanations algorithm tool to determine the impact of different input features on $Q_{\beta }$. Among these features, ${[\left(N\text{−}Z\right)/A]}^2$ is identified as the most significant factor. Unlike traditional symmetry energy terms which rank last, the ${[\left(N\text{−}Z\right)/A]}^2$ term ranks first and improves accuracy. This finding provides that the isospin chemical potential plays a more critical role in the $Q_{\beta }$ process.

Close

PrevNext

View figure in article

• Beta decay
• Binding energy & masses
• Electromagnetic transitions
• Data analysis
• Machine learning