For complex glass compositions with more than eight constituent compounds, experimental measurements of the entire composition space are prohibitively expensive and time-consuming. In addition, for systems with such complexity, there is no physically predictive model. There are requirements imposed on the end properties of glass and manufacturability requirements such as appropriate liquidus temperature and sufficiently low viscosity at a given temperature. These competing requirements necessitate the development of data-driven machine learning models of glass composition and properties. These models enable accurate and systematic prediction of glass properties such as Young’s moduli and liquidus temperature. Research companies with long track records of exploratory experimental research are in unique position to develop data-driven models by compiling and using their earlier internal experimental results. In this chapter, we present how Corning used this unique advantage for developing neural network and genetic algorithmic models for predicting compositions that would yield a desired liquidus temperature and Young’s modulus.