Accurate prediction of abdominal aortic aneurysm (AAA) rupture is essential because it has a high mortality rate when it occurs. Although hemodynamic indices are key predictors, conventional computational fluid dynamics (CFD) has high computational costs, making it difficult to apply in real-time clinical settings. In this study, we validated a deep learning framework, originally designed for thoracic aortic wall shear stress (WSS) inference, by applying it to anatomically complex AAA geometries. We generated a CFD training dataset based on longitudinal CTA data from 71 AAA patients. To optimize the model, we designed 14 experimental scenarios to evaluate the effects of input spatial resolution, neighborhood distance configuration, temporal data splitting to assess generalization, and synthetic Gaussian noise. Experimental results show that when combining high-resolution inputs (88×56) and short-range spatial features, the model achieves a low mean absolute error of 0.025 Pa and a high structural similarity index measure of 0.942. Additionally, the inference time was significantly reduced to less than 1 second, demonstrating higher efficiency. The pipeline proposed in this study suggests the clinical applicability of a real-time surrogate model for patient-specific WSS analysis and provides an input design strategy.