This dataset contains DFT-computed oxygen vacancy formation energies of different oxygen sites in over 1000 metal oxides present on the Open Quantum Materials Database (OQMD). Entries in the Data.csv file are indexed using the OQMD entry ID of the pristine structure, and list, alongside the oxygen vacancy formation energy, a number of other properties used as features in machine learning models. An exemplary code allowing to test the performance of a random forest regressor using different sets of features to predict the vacancy formation energy, as a function of training set size, is provided in Test_Models.py. More details on the dataset and machine learning models can be found in: arXiv:2309.01160. Acknowledgments The work was funded by the U.S. Department of Energy under Grant DE-EE0008089. S. G. acknowledges the Air Force Office of Scientific Research for support under Award No. FA9550-18-1-0136 (OQMD database). A. G. acknowledges the Center for Hierarchical Materials Design (ChiMaD) under Award No. 70NANB19H005 (ML models). A. J.A. S.-C. acknowledges the financial support to National Agency for Research and Development (ANID)/DOCTORADO BECAS CHILE/2018 - 56180024. A. J.A. S.-C. and T. L. acknowledge funding from the Toyota Research Institute through the Accelerated Materials Design and Discovery program (ML representations). The data was produced relying on the computing power provided by Quest high performance computing facility at Northwestern University.