The ultimate load of a stiffened plate is an important indicator in design and check, but the research is relatively scarce under transverse tensile loads. Therefore, this paper firstly conducted a systematic study using a tension experiment combined with digital image correlation techniques and a machine learning method. Preliminary results indicate that although machine learning methods can efficiently predict results that align well with the finite element model with shell elements, the prediction data tend to be larger than experimental data. To investigate the mechanisms for this phenomena, a detailed model of the failure of the laser-welded stiffened plates was conducted based on the established finite element model of ultimate load considering welding deformation, residual stresses, and material property degradation. The research results indicate that welding deformation and residual stresses weaken the structural load capacity. To achieve more accurate predictive results, the influence of both factors needs to be considered in machine learning models.