The oxygen evolution reaction (OER) is key for the transition to a hydrogen-based energy economy. The observed activity of the OER catalysts arises from the combined effects of surface area, intrinsic activity, and stability. Therefore, alloys provide an effective platform to search for catalysts that balance these factors. In particular, high-entropy oxides provide a vast material composition space that could contain catalysts with optimal OER performance. In this work, the OER performance of the AuIrOsPdPtReRhRu composition space was modeled using an experimentally obtained dataset of 350 nanoparticles. This machine-learned model based on experimental data found the optimal catalyst to be a mixture of AuIrOsPdRu. However, as a “black-box model”, it cannot explain the underlying chemistry. Therefore, density functional theory (DFT) calculations were performed to provide a complementary theoretical model with defined assumptions and, hence, a physical interpretation through comparison with the experimental model. The DFT calculations suggest that the majority of the activity originates from Ru and Ir active sites and that the addition of Pd improves the performance of these sites. However, the DFT calculation did not find the experimentally observed beneficial effects of Au and Os. Therefore, we hypothesize that the Os contributed to the performance of the tested catalysts by roughening the surface, whereas Au fulfilled the role of a structural support. Overall, it is demonstrated how machine learning can help accelerate catalyst discovery, and combining machine-learned models obtained from experimental data with models based on DFT calculations can provide important insights into the complex chemistry of OER catalysts.