The new approach being introduced uses the latest in machine learning techniques to more closely relate chemical spectrometry data to predicted physical properties, yielding better results than existing regression models. Several machine learning approaches were evaluated, including AdaBoost, Random Forest and MLP. After thorough analysis, the one with the highest R2 value and other useful characteristics was selected. It is seamlessly integrated into existing workflow and works in real time. The methodology is currently being extended to explore effects of other process variables on predicted physicals to present a more robust model.