<p>Welding technology continues to expand in various industries, resulting in an increasing demand for weldments characterized by high performance, precision, and reliability, which are often closely associated with residual stresses. Nevertheless, traditional experimental methods with diverse materials and process parameters are not only costly but also time-consuming. In this context, numerical simulations and machine learning modeling have emerged as viable solutions to tackle these challenges. Given their practical application value, this paper emphasizes the rigorously validated predictive models that have verified their practical applicability in real-world scenarios. A detailed review of the application of prediction models in the realm of welding residual stress is presented. Furthermore, this paper systematically evaluates the functionalities of various machine learning models, aiming to offer more precise guidance for constructing welding residual stress models. Additionally, in light of the existing knowledge gaps during the modeling process, we propose clear development trends and identify potential research avenues for future studies.</p>