Abstract: With the rapid advancement of radar technology, traditional inverse synthetic aperture radar (ISAR) image simulation methods based on electromagnetic (EM) scattering calculations often face the challenge of high time costs, making it difficult to generate high-resolution ISAR image samples in real time. To address the issue of low efficiency in constructing image sample datasets for complex targets, a machine learning-based model for fast ISAR image prediction is proposed. This model uses a small amount of ISAR echo data as the input for EM calculations of complex targets. Data augmentation techniques are employed to increase dataset diversity, and a dynamic weighted ensemble method is applied to integrate three commonly used regression models: linear regression, support vector machine (SVM), and random forest. The proposed ensemble model can rapidly predict ISAR echo data, reducing the number of EM simulations required and significantly improving the efficiency of sample generation. Experimental results demonstrate that the model can accurately predict all the data needed to generate images using only a small amount of echo data, achieving an overall efficiency improvement of approximately 80%. As the complexity and resolution of the target increase, the time required for simulation methods will increase significantly. At this point, our proposed model will demonstrate greater advantages.