Austenitic stainless steel is prevalent in the manufacturing of aircraft structural components and chemical pipelines, which are frequently subjected to high temperatures. To investigate the high cycle fatigue strength of welds involving two stainless steel materials S310 and S321, a series of high cycle fatigue tests were conducted at 550 °C, 650 °C, 750 °C, and 850 °C, respectively. Scanning electron microscopy (SEM) was used to observe the fracture surfaces in the base metal zone (BM), fusion zone (FZ), and heat affected zone (HAZ). The microstructure characteristics near the fusion line of the specimens were analyzed by the electron backscatter diffraction (EBSD). Results reveal a negative relavance between the presence of inclined coherent twin boundaries (CTBs) and material's fatigue performance. The initiation of cracks from the surface is caused by machining defects, welding voids, and high temperature oxidation of certain grains near the surface. For specimens that fracture in the FZ, the fatigue life show the divergence trend with temperature increases. Meanwhile, a fatigue life prediction model based on XGboost algorithm is developed with axial stress amplitude (σa), temperature (T), average grain size (Dmean), twin boundary density in the FZ (ρtb), and fracture location (Lf) as crucial analytical parameters. This model demonstrates better accuracy in predicting fatigue life with a mean absolute percentage error (MAPE) of 39.58 % compared with Basqin model's 46.81 %.