Background: The study of $\alpha$-decay half-lives is essential for understanding the stability and structure of heavy and superheavy nuclei. Accurate prediction of these half-lives is important for exploring regions where experimental data are limited, particularly in the superheavy domain. In recent years, machine-learning techniques have emerged as powerful tools for identifying complex relationships within the nuclear data and improving predictive capability.

Purpose: The present work aims to investigate the predictive capability of a neural network framework for estimating $\alpha$-decay half-lives across a wide range of nuclei and to examine whether the model can reliably reproduce known decay systematics and extend predictions to the superheavy region.

Methods: A neural network model was trained on experimentally known $\alpha$-decay data for 535 energetically allowed nuclei, comprising 159 even-even nuclei (${}^{148}\mathrm{Gd}$ to ${}^{294}\mathrm{Og}$), 295 odd-A nuclei (${}^{149}\mathrm{Tb}$ to ${}^{289}\mathrm{Fl}$), and 81 doubly odd nuclei (${}^{148}\mathrm{Eu}$ to ${}^{258}\mathrm{Md}$). The predicted logarithmic $\alpha$-decay half-lives are compared with the available experimental data as well as with the corresponding values obtained from the semi-empirical Universal Decay Law. The performance of the model is evaluated through correlation analysis, residual distributions, root-mean-square error behavior during training, and fivefold cross-validation tests. The trained model is then applied to predict $\alpha$-decay half-lives for theoretically known superheavy nuclei in the region 117 $\le$ Z $\le$ 120. The predicted results are further compared with recent machine-learning models, microscopic calculations, the finite-range droplet model, and the Viola–Seaborg–Sobiczewski formula.

Results: The neural network predictions show good agreement with experimental data and follow trends similar to those obtained from the semi-empirical formulas. The statistical analysis indicates high correlation between predicted and experimental half-lives, with small residual deviations and stable prediction errors across different data partitions. The model successfully captures the nonlinear dependence of $\alpha$-decay half-life on decay energy and nuclear charge. Predictions for superheavy nuclei exhibit systematic behavior consistent with established decay trends.

Conclusion: The results demonstrate that neural network techniques provide an effective and reliable approach for predicting $\alpha$-decay half-lives for even-even and odd-mass nuclei. The model reproduces known decay systematics and offers useful predictions in the superheavy region, which may assist future experimental studies of superheavy nuclei.

Close

PrevNext

View figure in article

• Alpha decay
• Binding energy & masses
• Nuclear structure & decays
• A ≥ 220
• Artificial neural networks
• Machine learning