<p>Vanadium slag-calcification extraction process is affected by variability in vanadium slag composition, which increases the technical challenges of vanadium recovery. These factors necessitate frequent experimentation to determine optimal recovery conditions, thereby increasing costs, time, and environmental hazard. In this study, a machine learning modeling approach was used to predict vanadium leaching rates based on previous research data for the vanadium slag-calcification roasting–sulphuric acid leaching process. By comparing multivariate linear regression, support vector machines, decision trees, random forest, and BP neural network models, the random forest model was found to perform best in predicting vanadium recovery rates, with an R² of 0.942 and an RMSE of 0.246. Furthermore, model analysis also revealed significant impacts of roasting temperature, leaching pH value, and n (CaO/V₂O₃) as the main factors influencing vanadium recovery rates. Finally, the optimal combination of feature parameters under the conditions of the highest vanadium leaching rate was obtained through optimization with the grid search algorithm, the reliability of the random forest model was verified through experimentation. This novel machine learning model not only creates new possibilities for efficient recovery of vanadium, but also reduced the risk of potential environmental hazards due to repeated optimization experiments, in addition, the entire process is safe and efficient.</p> Graphical Abstract <p></p>