Although thermodynamic analysis provides the possibility to perform a full-range study on the chemical vapor deposition (CVD) of SiC, billions of calculations are required to achieve a high-resolution study for such a high-dimensional space. Similar issues are also encountered for other materials modeling and simulation techniques. Therefore, using the database from the thermodynamic analysis of CVD of SiC as an example, we demonstrate a machine learning framework to accelerate physical-based materials modeling and simulations. Using only 6 % of randomly selected data generated in the thermodynamic analysis of CVD of SiC, a high-fidelity random forest (RF) with an accuracy (R2) of close to 1 was obtained. The derived key CVD metrics, i.e., deposit purity, deposition efficiency of SiC, and precursor utilization efficiency by the RF model are in excellent agreement with those determined by thermodynamic analysis.