Virtual metrology (VM) enables a fast and reliable methodology to track target outcomes of many prominent semiconductor manufacturing processes, especially in today’s sub-5nm node technology era. Chemical-mechanical planarization (CMP) is a microfabrication technique widely used in semiconductor manufacturing, typically multiple times, to achieve exceptionally planar surfaces. Virtual metrology factors in recipe setpoint knowledge, microfabrication equipment status information, and physical metrology data to construct a robust predictive model for the target variable(s). In addition to these factors, this study employs three important physics-based models to augment the traditional VM for CMP developed to predict the material removal rate. Notable attention is placed on the most important physical components of the CMP tool while selecting physics-based variables and their models, like the polishing pad, dresser, and the slurry being used. This study also relies on minimum redundancy maximum relevance (mRMR) technique to select a subset of all extracted features and rank them according to their relevance. Finally, a robust VM is developed using ensemble of trees, which yields an overall mean squared error of 13.71 nm/min2 on the newer data.