Machine learning algorithms, in conjunction with high-throughput atomistic simulations, are used to probe the uncertainty within the energetics, structure, and mechanical response of silicon carbide grain boundaries. In this work, a Metropolis style Monte Carlo algorithm is applied to tilt and twist grain boundaries and used to generate a large dataset of both stable and metastable structures. Tensile and shear response molecular dynamic simulations are performed on these grain boundaries, and using advanced structural descriptors, the initial configuration is related to the strength of these grain boundaries through the training of machine learning models. By considering metastable structures in addition to those in the ground state, this approach seeks to aid in the development of improved constitutive relations to connect microstructure and mechanical response. This would allow for the identification of critical flaws within the microstructure that limit the effectiveness of the material. Additionally, the machine learned relations could be used to atomistically inform higher length scale models, allowing for higher fidelity mesoscale models.