The exploration of low-lying excitation energies of atomic nuclei remains a pivotal area of nuclear physics research due to its unique capacity to uncover new physics phenomena, especially in regions far from stability. In this work, the state-of-the-art efficiency machine-learning algorithm Light Gradient Boosting Machine is applied for the first time to investigate low-lying excitation energies of $0_2^{+}$, $0_3^{+}$, and $3_1^{-}$ states in even-even nuclei. These particular excitations serve as crucial probes for quantifying shape coexistence and octupole deformation of nuclei. Our results are in good agreement with experimental data, achieving accuracy factors of 1.10, 1.12, and 1.07 for the $0_2^{+}$, $0_3^{+}$, and $3_1^{-}$ states, respectively. This represents a significant improvement over the predictions using the theoretical models calculations, as well as over previous machine learning approaches. The predictive capability of our machine learning methodology is further validated using 14 newly measured data points that are not used in the training set, demonstrating good agreement between the machine learning predictions and experimental data. Additionally, we encoded the significance ranking of input physics features relevant to the properties of low-lying excited states. The interpretable nature of present algorithm showcased robust performance, enabling timely and effective interventions that could enhance outcomes.

Close

PrevNext

View figure in article

• Collective levels
• Energy levels
• Nuclear structure & decays
• Artificial neural networks
• Decision trees
• Machine learning