The BP neural network improved by the dung beetle optimization(DBO) is used to calculate the stress concentration factor (SCF) of T-tubular joints, and the SCF can be solved quickly and accurately. First, finite element parameterized modeling of T-tubular joints under basic axial loading was conducted, and comparative analysis with experimental data verified the model’s reliability. Next, a SCF dataset was established for crown and saddle points, analyzing the influence of dimensionless geometric parameters on SCF. Finally, the BP neural network improved by DBO is used to perform regression prediction on the SCF data sets of joints with different geometric parameters. The results show that the prediction performance of the improved BP neural network model is better than that of the unimproved BP neural network. Compared with the SCF parameter equation, the BP neural network prediction using DBO is more efficient and accurate.