A naive Bayesian model averaging (NBMA) method is developed to predict nuclear masses. In the NBMA method, the weights of different models may be different for each nucleus, which are sensitive to the model accuracies to describe the nuclear masses of the isotopes and isotones with the same proton and neutron numbers of that nucleus. Therefore, there are remarkable local structures for the weights of different models on the nuclear chart, which well eliminates the local deviations between the model predictions and the experimental masses and thus achieves better accuracy of mass predictions than the traditional arithmetic mean method (AMM) and weighted average method (WAM). Based on the latest atomic mass evaluation of AME2020, the root-mean-square (rms) mass deviation of the NBMA method is 0.293 MeV, while the rms deviations of AMM and WAM are 0.634 and 0.361 MeV, respectively. This accuracy of the NBMA method is even 28% better than the best accuracy of the mass models used in the NBMA method. The extrapolation ability of the NBMA method is also verified with the experimental nuclear masses which are not used in the training of the NBMA method.

开发了朴素贝叶斯模型平均 (NBMA) 方法来预测核质量。在NBMA方法中，每个核的不同模型的权重可能不同，这对描述该核质子数和中子数相同的同位素和同位素的核质量的模型精度很敏感。因此，核图上不同模型的权重存在显着的局部结构，很好地消除了模型预测与实验质量之间的局部偏差，从而获得比传统算术平均法（AMM）和实验质量更好的质量预测精度。加权平均法（WAM）。根据AME2020最新的原子质量评估，NBMA方法的均方根质量偏差为0.293 MeV，而AMM和WAM的均方根质量偏差分别为0.634和0.361 MeV。NBMA 方法的精度甚至比 NBMA 方法中使用的质量模型的最佳精度还要好 28%。NBMA方法的外推能力也通过NBMA方法训练中未使用的实验核质量进行了验证。