Nonlocality Effect in the Tunneling of Alpha Radioactivity with the Aid of Machine Learning
Recently, building upon the research findings of E. L. Medeiros, we have extended the alpha-particle non-locality effect to the two-potential approach (TPA). This extension demonstrates that the integration of the alpha-particle nonlocality effect into TPA yields relatively favorable results. In the present work, we employ machine learning methods to further optimize the aforementioned approach, specifically utilizing three classical machine learning models: decision tree regression, random forest regression, and XGBRegressor. Among these models, both the decision tree regression and XGBRegressor models exhibit the highest degree of agreement with the reference data, whereas the random forest regression model shows inferior performance. In terms of standard deviation, the results derived from the decision tree regression and XGBRegressor models represent improvements of 54.5% and 53.7%, respectively, compared to the TPA that does not account for the coordinate-dependent effective mass of alpha particles. Furthermore, we extend the decision tree regression and XGBRegressor models to predict the alpha-decay half-lives of 20 even-even nuclei with atomic numbers Z=118 and Z=120. Subsequently, the superheavy nucleus half-life predictions generated by our proposed models are compared with those from two established benchmarks: the improved eight-parameter Deng-Zhang-Royer (DZR) model and the new empirical expression (denoted as "New+D") proposed by V. Yu. Denisov, which explicitly incorporates nuclear deformation effects. Overall, the predictions from these models and formulas are generally consistent. Notably, the predictions of the decision tree regression model show a high level of consistency with those of the New+D expression, while the XGBRegressor model exhibits deviations from the other two comparative models.