The mechanical properties of S-glass fiber reinforced epoxy composites are highly dependent on the fiber properties (i.e. strength, modulus and fracture toughness) and the surface reactivity between the glass surface and the polymer matrix (i.e. interfacial strength and energy absorption). Reactive atomistic scale simulations are useful to provide understanding of mechanisms among S-glass fiber, epoxy polymer, and amine molecules. For accurate calculation of S-glass and its interaction with sizing and matrix, parameter set for S-glass should be first developed. In this study, using a novel machine learning approach comprised of a Artificial Neural Network (ANN) assisted Genetic-Algorithm (GA), we parametrized ReaxFF interatomic potential parameters for Al/Si/O/Mg interactions for description of S-glass and other Magnesium Aluminosilicate (MAS) glass compositions (Figure 1). Our training set includes the Density Functional Theory (DFT) data of equation of state for various crystals of Mg/Al/Si/O such as Pyrope, Cordierite, Sapphirine, Forsterite, Enstatite, and Spinel. DFT based Nudged Elastic Band (DFT-NEB) calculation results for Mg migration inside Mullite crystal is also calculated and incorporated into the training set for better description of Mg ion migration. For validation, our new force field parameter set will be compared with experiment and quantum simulation data for mechanical properties and surface reactivity, such as Young’s modulus result and surface hydroxyl group number density. Our algorithm shows improved efficiency and speed for the ReaxFF training process over manual and GA only approaches. We expect this force field can be applied to construct a tool for virtual composition mapping for the development of new glass fiber materials. We also believe our force field would be able to support computational studies of amorphous materials used in geochemistry and construction material applications.