Document is current

Article History

Received: 26 August 2025

Revised: 17 October 2025

Accepted: 11 November 2025

First Online: 19 March 2026

Declarations

:

: A total of 1014 nuclei [ , ] were utilized in this study. The dataset was randomly partitioned into training and validation sets with an 8:2 ratio, which remained consistent throughout all subsequent training procedures. A separate test set consisting of 23 nuclei with charge radii measured since 2021 was employed for performance evaluation. The results summarized in Table demonstrate that the DNN model reduces the RMSE in predictions of charge radius by approximately 45% and 60% compared to the previous DNN model (DNN(i2)) and RCHB approaches, respectively. The computational implementation leverages the PyTorch deep learning framework. Robust convergence behavior was demonstrated, with the model typically reaching optimal performance within 10,000 training epochs. This training process required approximately 30 minutes of computation time when executed on an NVIDIA GeForce RTX 4060 GPU. The trained network exhibits remarkable inference efficiency, generating predictions in mere milliseconds—a feature that facilitates rapid deployment and real-time applications. The hyperparameters of the DNN are listed in Table  , and the dataset is attached at the end of the article.