In the rough rolling production process of strip steel, the width change is affected by the coupling of multiple factors. Accurate prediction of the width expansion is of great significance for improving the yield rate and product quality. This paper proposes a strip rough rolling width prediction model that integrates the mechanism model and the data-driven method. First, the width expansion mechanism model is established based on the Shibahara equation. After linearization by Taylor expansion, the least squares method and linear support vector regression algorithm are used to estimate the main mechanism parameters. Second, the extreme gradient boosting learning mechanism is introduced to correct the deviation after model parameter identification, thereby improving the robustness and accuracy of the prediction. Finally, the parameter convergence of the proposed method is verified in the simulation. The result shows that the proposed method is superior to the traditional modeling method in terms of prediction accuracy and convergence speed. This study provides a high-precision modeling idea for rolling width control that integrates physical constraints and data learning.