Aiming at the problems of high cost and long cycle in the fatigue strength assessment of metal structures under long-term loading histories, a new assessment method utilizing machine learning technology was proposed. The new method comprises three steps: structural fatigue modification, equivalent load calculation, and fatigue strength assessment. By taking four-point bending components and hole-containing plates as examples, comparing the fatigue strength utilization rates obtained from results without machine learning and those derived using the machine learning method, and by analyzing the influence of sample size on the model’s accuracy, the new method was verified to have the advantages of high accuracy and high efficiency. It is proven that this method is suitable for the fatigue strength assessment of metal structures subjected to long-term loads and multiple working conditions and significantly improves computational efficiency and reduces costs. Furthermore, the method demonstrates good generalizability, making it suitable for the fatigue strength assessment of different structures and holding considerable engineering value.