Predictive models of material properties trained using multi-fidelity data, A strategy for machine learning has been developed that exploits the fact that data are often collected in different ways with varying levels of accuracy. The approach was used to build a model that predicts a key property of materials.